IntroductionThe advancement of technology has made a significant difference in the healthcare field, particularly in providing open-handed solutions to people with disabilities. As a result of paralysis, the paralyzed face problems that require appropriate and innovative treatment options as well as adequate monitoring. The correct posture mitigates several health complications in such people, including pressure sores, muscular contractures, and respiratory problems1. These unsuitable sitting postures have been found to increase a person’s risk of sliding out of a wheelchair and causing further injuries. It is, therefore, important to closely monitor these individuals’ postural and health status to enhance their quality of life. A new framework, Emperor Penguin optimized sensor-infused Wheelchair (EPIC), can help mitigate these challenges by integrating advanced sensor technologies with optimized Deep Learning (DL) models.Due to their inability to move or turn independently, paralyzed patients face many challenges. Many complications can arise from immobility, most of which can be prevented with reasonable care2. Due to continuous pressure on body parts, the development of pressure ulcers, or bedsores, is of immediate concern. It may result in a partial lack of blood circulation to the affected areas, causing tissue damage without periodic mobilization. An incorrect position can also lead to muscle contractures, in which muscles shorten irreversibly, making it difficult to move and altering the shape of joints. It is also predicted that people with poor posture will have respiratory problems due to squashed lungs that cannot expand properly, which limits the amount of oxygen they can breathe, further complicating breathing problems3.These are just a few problems highlighting the need to implement continuous monitoring systems so caregivers can react to complications before they become severe. Healthcare Monitoring Systems (HMS) typically require periodic manual attention, which is time-consuming and prone to human error. As a result, checking the posture and condition of such a paralyzed patient requires considerable physical and mental effort from the caregiver. Due to the effort involved, it is necessary to have automated monitoring and alert systems in place to ensure adequate care for such a paralyzed person. Technological advancements have recently allowed us to construct real-time paralysis health and posture monitoring systems based on sensor-based technologies and Artificial Intelligence (AI).Sensor technology has significantly advanced in these technologies, facilitating accurate and continuous data collection from the human body. In addition to monitoring pressure distribution, body orientation, and movement patterns, these sensors can be integrated with multiple devices, such as a wheelchair. In contrast, AI algorithms analyze, make decisions, and forecast results from a large volume of data through identifying and predicting patterns in the data. As a result of Machine Learning (ML), healthcare gains the ability to diagnose diseases, predict patient outcomes, and personalize treatment plans4. Nevertheless, the effectiveness of such algorithms is determined mainly by the quality of the data provided. As a result, relevant feature selection algorithm plays a significant role in its development5.A paralyzed patient can access all monitoring dimensions with the EPIC framework, which combines sensor technology with artificial intelligence. Several components comprise the system: sensors embedded in the wheelchair, an Arduino board to collect data, an ESP8266 Wi-Fi module to transmit data, and a Deep Maxout Network (DMN) to predict postures. The Emperor Penguin Optimizer Algorithm (EPOA) is also integrated with the framework for feature selection. There are two types of sensors used in the proposed EPIC framework namely Force Sensitive Resistor (FSR) and ultrasonic sensors. In these sensors, readings are always collected concerning the pressure of the body of the discrete against them and the relative positions of the body. Data is collected by these sensors and sent to an Arduino board for processing before being sent to an ESP8266 Wi-Fi module. When this data is received from the sensing layer, the Wi-Fi module sends it to a central server so the EPOA can process it.A key objective of the study is to determine the most appropriate feature selection algorithm based on the data collected within EPIC. As part of an ML process, feature selection is one of the most critical steps, directly affecting the model’s performance. In this paper, an feature selection algorithm, EPOA, is presented inspired by the foraging behavior of Emperor Penguins. This model attempts to identify the most significant features contributing to its accuracy. The selected number of relevant features is then input into the DMN. Through this process, the DMN learns to predict the posture of a wheelchair-using individual based on the input features. The DMN output is then sent to an Android application to visualize the patient’s predicted posture and health in real-time. As a result of this method, the caregiver can monitor the paralyzed person’s condition constantly, and corrective measures can be taken immediately if problems are detected6.Compared to the traditional HMS, the proposed EPIC framework offers several advantages. As part of the model, continuous and real-time HMS is proposed, which can prevent further complications in paralyzed patients7. By integrating advanced sensors, accurate and comprehensive data will be collected, including the patient’s posture and health status. By using EPOAs as feature selection algorithms, the system’s accuracy can be improved, ensuring only the most relevant data is used in predictions. Furthermore, the DMN can capture complex nonlinear relationships between multiple factors, allowing better and more relevant posture predictions to be made. It becomes vital in healthcare, where even a tiny error may have serious consequences8. By providing caregivers with real-time alerts and updates, EPIC reduces the risk of complications and improves quality care for the paralyzed.Moreover, an Android application displays the predicted posture status and health conditions, making the system easy to use and accessible. Caregivers can know the condition’s status without being physically present, which reduces the need for constant physical presence so they can do other things9. This will not only enhance efficiency in care but also reduce the burden on caregivers, making the EPIC framework a vital tool in managing paralyzed individuals.The main goal of this work is to develop a complete system that monitors posture and checks health in real-time for people who can’t move. This system uses the new EPIC. The plan will use pressure and ultrasonic sensors to collect data about the user’s posture and health. The system uses intelligent methods like the Emperor Penguin Optimizer (EPO) to select essential details and a DMN to predict posture. The system also sends this key data to a mobile application so caregivers can monitor the user’s condition as it happens10. Ultimately, the system aims to stop health issues like bed sores, muscle tightening, and breathing by maintaining good posture. This should make life much better for people who use wheelchairs.Key contributions of the article are to track and analyze user posture using FSR and ultrasonic sensors, ensuring continuous assessment to prevent complications like pressure ulcers and muscle contractures. Also, an advanced feature selection approach is introduced using the EPOA efficiently extracts and refines relevant data from multiple sensors, optimizing computational efficiency and reducing redundancy. Further on, the deep learning-based posture classification model developed using the DMN enhances recognition accuracy by effectively handling non-linearity in sensor data, outperforming existing techniques. Based on these, an IoT-enabled real-time alert system is developed that transmits posture data to an Android-based mobile application, allowing caregivers to monitor wheelchair users remotely and receive instant notifications about critical postural deviations11.The rest of the article is organized as follows: Sect. 2 highlights the existing methods from the literature, which was reviewed, and research gaps are identified. Section 3 describes the EPIC framework, including how data is acquired and processed and how posture predictions are made using the DMN. Section 4 presents the experimental results, compares them with existing techniques, and discusses the results. Finally, Sect. 5 summarizes the work and suggests future directions.Related worksThis section focuses on current studies on posture observation, sensor-integrated wheelchair support, and deep learning methods in healthcare solutions. Its review points out challenges with the current process and stresses the need for a better, faster, and more flexible system such as the EPIC framework.Lately, designing practical algorithms to optimize several components of Wireless Sensor Networks (WSN) and HMS has been a flourishing research domain. This section overviews recent developments in optimization techniques and their applications across diverse domains to enhance network performance, security, and resource efficiency. Towards this effort, the paralyzed human monitoring with an innovative approach can be done based on the sensor-infused EPO-DMN. Featuring sensor devices enables the development of a holistic system with optimized algorithms involving the EPO and the DMN. The developed system is capable of detecting events through image processing and, in an efficient manner, sending data and correctly classifying tone-mapped images for visualization applications. The expanded network can also balance load and minimize energy in multi-hop communication. The objectives will be achieved by integrating sensor technology, optimization algorithms, and DL models to make a robust HMS that effectively supports people affected by paralysis and ameliorates their quality of life12.Wireless sensor networks (WSN)An efficient user association in mmWave networks can improve throughput maximally, ensuring spectral efficiency. The contribution of the EPOA to mmWave network user association schemes has been identified as immense. In this respect, the EPOA ensured a maximum throughput of 3.2 × 109 bits/s and, with very high spectral efficiency of 80 Mbps/Hz, surpassed state-of-the-art results in both measures. EPO and EPOA are closely related but serve distinct roles within the EPIC framework. EPO is a bio-inspired optimization technique modeled on the huddling behavior of emperor penguins, where individuals dynamically adjust their positions to maximize warmth and survival. EPOA refers to the specific implementation of EPO within the EPIC framework, customized for feature selection in sensor-based posture monitoring.This comes with reduced complexity and cost and thus forms an attractive solution for next-generation WSN13. The integration of Ultra-Wideband (UWB) radar systems in person monitoring applications has been recognized to realize their possible applicability, particularly in dense indoor environments. The ability to operate with an exemplary precision hardware demonstrator for real-time, wireless UWB Sensor Network (UWB-SN) was developed to help keep track of the movement of human beings, especially with higher resolution requirements. The system has a modular architecture that allows for fast development and fielding of tracking applications. In this M-sequence UWB radar front-end implementation, low-rate data streams of compressed target coordinates are generated and sent to a fusion center for processing. This approach improved the tracking capabilities and confirmed the real-time performance of UWB-SN hardware, demonstrating acceptable precision even using 32-bit arithmetic in the software algorithms process14. Accurate feature selection is important for enhancing classification tasks over wireless sensor networks. In this line, it has been evidenced that MPSA enhances the accuracy of feature selection but reduces the processing time. It has been exposed that it can be applied efficiently to environmental monitoring tasks, in which the MPSA eliminates redundant features to increase the accuracy of the classification. This improvement is precious within forest environments, where variables like temperature, humidity, light intensity, and carbon dioxide levels must be recorded across different climatic zones, including morning, afternoon, and night15.One of the critical requirements in HMS is detecting important events and efficient storage of data related to events. Modern sensor devices are equipped to detect prescribed events related to a monitored person and store event-related moving pictures for an extended period. These systems generate reports to central management apparatuses upon detecting these prescribed events. In these systems, event preservation and detection units effectively manage large data quantities, maintain vital information, and make it referable in the future16.Deep neural networks (DNN)Security in Wireless Multimedia Sensor Networks (WWSNs) is becoming more paramount due to the interest in the intrusion of Denial of Service (DoS) attacks. One optimized Deep Neural Network (DNN) is now proposed for the same purpose using parameters selected through an Adaptive Particle Swarm Optimization (APSO) algorithm. That enhances the efficiency of DoS attack detection and diminishes the computational overhead linked with traditional approaches. The optimized DNN detects DoS attacks impeccably and is robust in securing WMSNs. A DoS attack may result in delays or loss of sensor data, hindering prompt alerts for caregivers and risking patient safety. The framework could be integrated with lightweight authentication methods to guarantee the system’s robustness, dependability, and ability to maintain uninterrupted monitoring without outside disruption. Recently, a new algorithm called HIGHT, in the realm of cryptography, has emerged, and it tries to develop an algorithm that is much more friendly in implementations than AES on small devices with harsh resources, like RFID tags, sensor nodes, and smart cards. Hence, it delivers an advanced software implementation that eases the computational load on these appliances. Therefore, such work is likely appropriate for applications requiring both speed and keeping the chip size to a minimum. In particular, this makes efficiency very important for all devices operating under challenging conditions, such as limited power and processing capacity17,18.Sensor networks for real-time environmental monitoring have been in their design phases, guided by optimization frameworks in attaining optimal network performance, ensuring that all operational demands are met by the network structure: one that optimizes network layout in efficiently satisfying these demands, thereby availing that the WSN works effectively in some mixed environments. However, this critical optimization in applications requiring continuous monitoring lacks detailed study, for example, monitoring repeated changes in temperature, humidity, and other environmental factors19. For applications in an industrial setup, the WSN design typically requires optimizing the processes in play to monitor effectively. Independent Component Analysis (ICA) was recently employed in this regard, showing its ability to enhance the performance of the process monitoring system with optimal accuracy in the results while lowering the computational load. Here, ICA accelerated the designing of sensor network deployments that would enhance the overall control and efficiency of processes by accurately improving monitoring and optimizing the general approach for deploying sensors20. The optimization of the Long Short-Term Memory (LSTM) network by the Penguin algorithm has been seen to make actual, honest improvements to the Intrusion Detection System (IDS). It has led to high accuracy rates on datasets, e.g., NSL-KDD, where the training accuracy was 99.4% with a testing accuracy of 98.8%. The NSL-KDD dataset serves as a commonly utilized benchmark for assessing IDS. It enhances the KDD’99 dataset by resolving issues related to repetitive and duplicate entries, which had caused biased learning in machine learning models. This dataset offers a more balanced view of normal and attack traffic, facilitating improved generalization in security-focused applications21.These results have shown that the LSTM network optimized by a Penguin algorithm is very efficient in detecting and prohibiting unauthorized access in networked environments. This means that a better coverage rate offered by the sensing nodes in a WSN form is critical when full area coverage is required in specific applications22. The optimized bat algorithm was used to attain higher coverage rates than other algorithms, such as the Flower Pollination Algorithm (FPA). This optimization raises the mean coverage rate further, and besides reducing computational time and standard deviation, the network deployment becomes more stable and efficient. Other fields where WSNs are implemented are monitoring the conditions of utility poles. The sensors embedded into utility poles detect any physical change and transmit data to a remote server whenever necessary. This ability ensures that any degradation of the pole’s condition is noticed and reported so that poles can be effectively maintained before hazards are caused23.This has prompted the realization of an auto-decorrelation block for the Transformer models in industrial multi-sensor data processing, which reduces memory usage to 1/5th with comparable accuracy and cuts the inference time in half. This makes the model suitable for real-time applications where speed and efficiency are paramount24. Particularly in the processing of videos in sports, event recognition and summarization have been substantially enhanced using the Hybrid Deep Neural Network (HDNN)-EPO method. It has been applied in the examination of cricket videos, where it performs comparatively better than the existing techniques used to recognize events. Extracted highlights compare positively well with highlight videos prepared manually based on human decision, showing effectiveness in capturing the live parts of sports videos based on HDNN-EPO25.Real-time continuous HMS depends on the uninterrupted data transmission made by these sensors to the end user about a patient. An interfacing device was developed that allowed the health monitors to be interfaced with the various sensors. They supported 2 of the data transfer modes. This enables the integration of several types of sensors. A total monitoring of temperature data, electrocardiogram (ECG), individual sound, respiration, and blood pressure can be integrated from this system of the respective patients. This means that efficient data transmission from the sensors to the monitors is an existential issue; it supports or determines the accuracy and reliability of the HMS26. This literature survey emphasizes tremendous progress in optimizing WSNs, algorithms, and systems for various applications, ranging from health and environment to safety and industrial process control. Each of the innovations thus contributes to developing more efficient and accurate systems that are reliable for meeting the growing demand for modern technology27.Sensor network design for process monitoring is paramount nowadays in industrial applications. In application to synthetic data sets and the real-world data process, the proposed technique worked much better in the WSN design than in previous studies. The ICA technique enhances the process monitoring of improved industrial practices to a greater extent. By incorporating ICA, the accuracy in monitoring processes comes out very high; hence, an optimized sensor network design that supports better performance and reliability in the vast arena of different industrial environments will be set forth19. It is observed that the security domain of the networks has brought immense improvements using optimization-based LSTM networks coupled with the EPOA in attending to numerous IDS20. This has optimized accuracy, efficiency, and adaptability in IDS. From the dataset, the optimized LSTM network is up to an astonishing 99.4% on the training dataset, whereas the test dataset goes up to 98.8%. At the same time, exceeding the existing practical techniques proves the efficacy of the implementation in securing environments. Coverage rate optimization is a key aspect of implementing the WSN.Optimizing the bat algorithm improved the coverage rate when compared with the FOA. This optimization enhances the coverage rate and reduces the overall number of the remaining nodes in the network. The simulation results show that the optimized bat algorithm performs much better than FOA in the mean coverage rate, computation time, and standard deviation. It is a superior node-deployment approach for WSN28.Posture monitoring techniquesDhiman et al.29 have proposed a DL and optimization-based framework called DON to identify the novel coronavirus disease through X-ray images. Although it emphasized medical image processing, the research indeed screened the application areas of DL models for their potential in the health sector and other application domains, such as posture-monitoring applications. The authors proposed a hybrid PolyNet model based on clinical fundus images to estimate the glaucoma severity level30. This research attests to the necessity of hybrid models, merging DL solutions with other computational approaches, for the effective and accurate diagnosis of many diseases. In this context, it is relevant to posture monitoring, and it can enable such hybrid models to perform better, especially in classification scenarios. Following this, we reviewed the use of optical fiber sensors for posture monitoring, ulcer detection, and control for wheelchairs. The study emphasizes the state-of-the-art condition of sensor technologies applied to wheelchair applications, including important implications for the accurate achievement of sensor data in order to satisfy the requirements of posture monitoring effectively31. Hou et al.32 developed a prototype autonomous wheelchair with an implanted HMS based on the Internet of Things (IoT). The work shows that it is possible to integrate IoT into wheelchairs to provide real-time HMS; hence, it acts as a proof of concept for the blueprint related to the proposed DMN framework with IoT capability.Bhaskar et al.33 contributed a real-time remote-controlled wheelchair with Wi-Fi interference, including obstacle detection for disabled and elderly persons. Their research outlooks the elevation into improved maneuverability and safety of wheelchairs through data communicated in real-time in this manner, the intended goal resonates strongly with the envisaged framework in terms of monitoring user posture and real-time response. Where the works of34 presented intelligent systems concerning sitting postures and anomaly detection in 2024, this section categorizes different approaches and technologies used in posture monitoring systems, which gives a clear insight into the pros and cons of existing methods35. showed their system, a voice-controlled wheelchair for health monitoring with oxygen cylinder integration.It highlighted the increasing trend to integrate multiple assistive technologies within one system and is relevant to the more comprehensive development of a posture and health monitoring solution36. developed an IoT-enabled moving wheelchair with obstacle detection and continuous HMS. The work emphasized the methodology of constant monitoring and real-time feedback, which remains the cornerstone of the proposed framework.Based on this literature survey, optimization techniques were outlined in the area above. The integration of these optimized algorithms and methods in the following sections with applications has since witnessed noticeable performance, efficiency, and reliability enhancements in current practice, paving ahead for much more robust and scalable systems. The proposed EPIC framework is a significant improvement in populating the postural and health conditions of paralyzed individuals, although there are some research gaps. These consist of real-time adaptability to each user’s needs, integrated HMS besides the postural part, and diversified sensors to guarantee higher data accuracy. Power efficiency and battery life, making it more interactive by implementing interfaces that improve the caregiver’s interaction and further validation across multiple subjects, exist.Moreover, integration of the EPIC framework with other assistive or rehabilitation technologies to suggest a comprehensive and effective solution for paralyzed individuals is pending. What is most demanded by the design of WSNs optimized for real-time environmental monitoring is the fact that the operational requirement is satisfied while network performance is maintained. For all such purposes, the network layout should be optimized to the greatest possible height to maximize the network’s performance. The WSN designed using this framework shall ensure real-time effectiveness in carrying out a HMS of all environmental variable parameters. The better design and layout of sensor networks will translate to further contribution of the said framework in realizing more reliable and accurate ecological HMS.Recently, it has been a subject of interest for intelligent systems to be further advanced while contributing a more expansive mechanism of potential benefits that Artificial Intelligence (AI) promises to wheelchair users with walking disabilities. Some studies have also been conducted in the field of developing intelligent systems for posture monitoring, health monitoring, and autonomous navigation in wheelchairs. All those research efforts have provided a very sound footing for the proposed HMS based on the DMN.Methods and materialsThis section outlines the incorporation of sensors, data handling, and machine learning methods employed in the EPIC framework for monitoring posture in real time. It includes the hardware setup, feature selection using EPOA and DMN categorization, and data transmission through IoT to notify caregivers.Pressure sensorsThe data range of pressure sensors extends from 0 to 1023, typical for standard analog sensors interfaced to an Arduino or other microcontrollers. This information aids in identifying how a user’s weight is distributed over a seating surface, which is crucial for analysing potential posture issues and confirming proper seating. Figure 1 illustrates the data visualization corresponding to pressure and ultrasonic sensors. Figure 1a presents a scatter plot that shows the pressure sensor readings collected from various FSRs in the wheelchair seat. We designed these sensors to observe the weight distribution and ensure posture stability. We methodically arrange sensors labeled s1, s2, and s3 to detect pressure differences between different areas of the body while the individual is sitting. The plot’s axes denote the sensor values as ADC readings, reflecting the pressure detected by each FSR. We can convert these ADC values into force (N) or pressure (Pa) using the sensor’s calibration information to enhance understanding. The graph shows the interconnections between s1, s2, and s3, helping to examine the pressure change relationships at designated seating locations. Figure 1b displays the voltage fluctuations observed in the ultrasonic sensor, depicting more straightforward distance measurements to analyze. A 5 V reference and a 10-bit resolution convert the raw ADC values to do this, which ensures that posture assessments are correct and allows real-time tracking within the EPIC framework.Fig. 1Visualizing the data in (a) pressure sensors and (b) ultrasonic sensors.Full size imageUltrasonic sensorS4, the range 0–15 cm, to measure the distance between the back of the user and the backrest of the wheelchair. This information will help find that the user is leaning too much backward or forward, which may indicate a problem with the posture. Figure 2 The histograms represent the values spread for each sensor (s1, s2, s3, s4). The dataset can be downloaded from https://ieee-dataport.org/open-access/data-set-wheelchair-sensors.Fig. 2Histograms showing the distribution of values for each sensor (s1, s2, s3, s4).Full size imageThe chosen HC-SR04 ultrasonic sensors operate within a range of 2 cm to 400 cm, providing comprehensive coverage for detecting the alignment of backrests and footrests. These sensors are installed at a 45° angle for backrest monitoring to observe spinal positioning, and at a 90° angle for footrest alignment to evaluate lower limb posture. To reduce the influence of mechanical vibrations from wheelchair movement, an adaptive filtering mechanism is utilized, which dynamically refines threshold values and employs Kalman filtering to minimize noise in distance measurements.Proposed methodologyFigure 3 shows a wheelchair posture monitoring system. Using sensors, the system acquires and processes data to determine whether the user has assumed a safe posture. The information might be transferred to a smartphone application, which can provide timely feedback or alert caregivers if necessary. It has several elements: sensors, processing units, and smartphone applications. Additionally, the diagram shows the flow of data in the system. The multiple sensors will collect data regarding the user’s posture. The force-sensitive resistor sensor measures the pressure the user applies to the wheelchair. It provides information about the weight that is placed on the wheelchair seat. In this wheelchair, a DC motor is used to control the movement. In addition to moving forward, backward, and sideways, the wheelchair can be turned with this motor’s help. Ultrasonic sensors calculate the distance between the wheelchair and objects around it. As a result, collisions with objects and walls are avoided.Fig. 3Architecture of the proposed workflow.Full size imageInitially, the assessment was centred on static posture analysis. Still, now we examine the effects of variations caused by movement, such as changes in pressure distribution and sensor alignment due to acceleration, deceleration, and uneven terrains. The FSR sensors adapt dynamically to variations in pressure patterns, while the ultrasonic sensors include a real-time correction mechanism to address vibrational noise and small misalignments resulting from movement. Moreover, an adaptive filtering algorithm has been applied to stabilize sensor fluctuations, guaranteeing precise posture classification even while the wheelchair is in motion.The processing unit analyzes the sensor data. An algorithm determines whether the user is in a safe position. A DMN forms the algorithm. The EPOA has been used to optimize this neural network. The processing unit sends the data to the smartphone application, which depicts it to the user. It can also generate alert signals to caregivers in case of a user’s dangerous posture. Figure 3: Flow of data through the system. The sensors are used to collect data about the user’s posture, and the data is then sent to the processing unit. The processing unit analyzes the data and determines whether the user is in a safe posture. The processing unit sends this information next to the smartphone application and then displays it to the user. This work recommends a multiple-stage methodology for monitoring the posture and health of paralyzed individuals using the EPIC framework (Fig. 4).Fig. 4The flowchart of the proposed EPOA.Full size imageData acquisition via embedded sensorsThe system starts with the continual collection of data through embedded sensors that are integrated into the wheelchair.There are two significant types of sensors used:(a)Force Sensing Resistors (FSR): These types of sensors compute the pressure the patient’s body exerts on numerous parts of the seat of a wheelchair, thus providing essential information about their posture.(b)Ultrasonic Sensors: These sensors will further detect the position and distance of the patient relative to the surfaces of the wheelchair, informing the posture analysis.Mathematically, let \(\:{S}_{FSR}\left(t\right)\) and \(\:{S}_{US}\left(t\right)\) represent the data streams from the FSR and ultrasonic sensors at the time ‘t’, respectively. These data streams are sent to the Arduino board for initial processing, EQU (1)$$\:{S}_{Data}\left(t\right)=\left\{{S}_{FSR}\right(t),{S}_{US}(t\left)\right\}.$$(1)Sensor fusion stageTo classify posture, the EPIC framework integrates data from FSR and ultrasonic sensors into a unified dataset and transmits it to the DMN. This integration optimizes pressure distribution data from FSR sensors and spatial posture data from ultrasonic sensors. As a result, it enables a more precise assessment of posture. For effective feature fusion, we employ EPOA to select and integrate features. The initial stage of this process involves preprocessing and normalizing the data. We used Min-Max scaling to align the raw data from the FSR and ultrasonic sensors onto a standard scale. This ensures that the results are the same for all sensor types. After normalizing the data, the EPOA-based feature selection process begins. The optimizer looks at sensor data using entropy measures and fitness scores, removing unnecessary or essential features for classifying posture and keeping only the most important ones. This process not only improves computational efficiency but also preserves vital information. Once the features are chosen, better pressure distribution features from FSR sensors are mixed with spatial posture features from ultrasonic sensors to create a single feature vector. A weighted approach makes the fusion even better, giving more weight to FSR features because they are more closely linked to the pressure imbalances that cause bad posture. After that, the DMN takes the combined feature vector and uses its unique activation function to effectively capture hierarchical feature representations and handle nonlinearity.Data transmission stageThe data from the FSR and Ultrasonic sensors are sent to an Arduino microcontroller for initial processing. This Arduino is an edge central processing unit that merges the sensor data into a single stream. In mathematical form, data stream, EQU (2).$$\:{S}_{Data}\left(t\right)=\left\{{S}_{FSR}\right(t),{S}_{US}(t\left)\right\}$$(2)where:\(\:{S}_{FSR}\left(t\right)\:\)represents the pressure data from the Force Sensing Resistors at the time ‘t’.\(\:{S}_{US}\left(t\right)\) represents the distance data from the Ultrasonic Sensors at the time ‘t’.The Arduino processes this combined data stream to ensure that the sensor readings are synchronized and that any initial filtering or noise reduction is applied. This preprocessing step is crucial for ensuring the data is reliable and ready for further analysis.The collected sensor data, \(\:{S}_{Data}(t\)), is transmitted to an Arduino board, which is a central hub for initial data processing. The Arduino board pre-processes the sensor data to ensure it is in a format suitable for wireless transmission. This pre-processed data is then sent to the ESP8266 Wi-Fi module, EQU (3)$$\:{S}_{ESP}\left(t\right)={f}_{Arduino}\left({S}_{Data}\right(t\left)\right)$$(3)where \(\:{S}_{ESP}\left(t\right)\:\)is the data ready for wireless transmission, and \(\:{f}_{Arduino}\) represents the pre-processing function performed by the Arduino board.Each analogRead() function returns a value between 0 and 1023, corresponding to the voltage level at that pin. For FSR sensors, this value indicates the amount of pressure applied. For the ultrasonic sensor, the value typically represents the distance between the sensor and the object in centimeters. This step is crucial for packaging the sensor data into a single-string format that can be easily transmitted or stored. The resulting string SData can be sent over a communication interface (Wi-Fi) to another device for further processing.After the initial processing by the Arduino, the data stream \(\:{S}_{Data}(t\)), is transmitted to an ESP8266 Wi-Fi module. The ESP8266 acts as a communication gateway, enabling the data to be sent via Wi-Fi to the central server or cloud for further processing.The transmission of data can be done via a serial connection between the Arduino and the ESP8266 module. The data stream \(\:{S}_{Data}(t\)), is then encoded and sent over Wi-Fi to ensure the real-time monitoring system remains responsive and efficient. The code is designed for an ESP8266 microcontroller to connect to a Wi-Fi network and transmit sensor data to a server. The ESP8266 first establishes a Wi-Fi connection using the provided SSID and password. Once connected, it continuously listens for incoming serial data from sensors or another device. When data is received, it packages it into an HTTP-POST request and sends it to a specified server. This setup is commonly used in IoT applications to send real-time sensor data to a remote server for processing, storage, or analysis.Feature selection using EPOOnce the data is transmitted to the ESP8266 module, Wi-Fi is sent to a server or local processing unit where feature selection is performed using the EPOA. The EPOA is key for identifying the most relevant features from the sensor data that will be used for posture prediction, EQU (4)$$\:{F}_{Selected}=EPO\left({S}_{ESP}\right(t\left)\right).$$(4)Here, \(\:{F}_{Selected}\) represents the set of feature selection by the EPOA from the sensor data \(\:{S}_{ESP}\left(t\right)\).The objective function for the EPOA can be mathematically represented as Equ. (5)$$\:Maximize\hspace{0.17em}J\left(F\right)=\sum\:_{i=1}^{N}{w}_{i}.{f}_{i}\left({S}_{ESP}\right(t\left)\right).$$(5)Subject to Eq. (6)$$\:\sum\:_{i=1}^{N}{w}_{i}=\text{1,0}\le\:wi\le\:1.$$(6)where \(\:{f}_{i}\) represents the individual feature functions, \(\:{w}_{i}\:\)does the EPO algorithm assign the weights, and J(F) is the objective function representing the overall effectiveness of the feature selection. The EPO is a nature-inspired algorithm based on the huddling behavior of Emperor Penguins in extreme cold. Penguins huddle together to minimize heat loss, and this behavior is modelled to solve optimization problems. Algorithm 1 shows an outline of the EPOA and how it is used in the proposed work:Algorithm 1 An outline of the EPOA.Here, an EPOA has to be used for feature selection from the sensor data, namely pressure and distance measurements. The feature selection is significant in reducing the dimensionality of data, which will benefit the performance of the subsequent prediction model. Only relevant features that contribute to accurately predicting posture output in a patient who uses a wheelchair will be identified through the EPOA. This supports reducing computation time and improving model accuracy. After feature selection, the most significant features are fed into a DMN to predict the patient’s posture. The EPO guarantees that only substantial features will be used, thus improving the expected performance of the DMN.Posture prediction using DMNThe selected features \(\:{F}_{Selected}\) are then passed to a DMN for posture prediction. The DMN is a type of neural network that enhances prediction accuracy by using the Maxout activation function, which is particularly effective in handling non-linearities in the data, EQU (8).$$\:{P}_{Predicted}=DMN\left({F}_{Selected}\right)$$(8)where \(\:{P}_{Predicted}\) represents the predicted posture of the patient who uses a wheelchair.The Maxout activation function is defined as Equ. (9)$$\:f\left(z\right)=\underset{i\in\:\left[1,k\right]}{\text{Max}}({z}_{i})$$(9)where ‘z’ represents the input to the activation function and ‘k’ is the number of linear bits in the Maxout function.Algorithm 2 For using DMN based posture prediction.First, the EPO is applied to the raw sensor data represented by pressure and distance readings, followed by FS. The DMN consists of several layers, among which are max-out layers transforming the feed input into a form that will be easier to classify. It processes the input features to produce an ultimate continuous prediction of the patient’s posture, which can be sitting upright, leaning forward, or sideways, and the prediction is sent to an HMS to inform caregivers about a patient’s posture in real time.The feature selection by the EPO would form the input to the DMN, which may include but is not limited to, pressure distribution and distance measurements through sensor readings. Weights and biases are randomly initialized for the DMN, learned during the training phase. The data is then forwarded through the network in which every layer computes a linear combination of the input followed by the maxout activation function, as shown in Fig. 5.Fig. 5Deep maxout network.Full size imageThe maxout function picks the highest value from several linear combinations in each layer. This allows the network to shape complex nonlinear interactions between features. The data then moves through multiple hidden layers. The final layer uses a SoftMax function to create a probability distribution for the possible posture classes. The system chooses the posture class with the highest probability as the patient’s predicted posture. Caregivers then get this prediction to learn about the patient’s posture in real time.Table 1 Hyperparameter tuning.Full size tableTable 1 outlines the hyperparameters explored during tuning, listing potential values and the final selected value for each to optimize the DMN’s performance.Display and alert through android applicationThe predicted posture data, \(\:{P}_{Predicted}\), is transmitted to an Android application via an ESP32 module. The application displays the patient’s real-time posture and health status to the caregiver. The application also alerts if the patient’s posture deviates from a safe position, ensuring timely intervention, Equ. (10).$$\:{App}_{Display}\left(t\right)={P}_{Predicted}\left(t\right)$$(10).System implementation detailsHardware specificationsMicrocontroller: Arduino Uno is a central processing unit for acquiring and processing sensor data from the wheelchair.FSR: Three FSR sensors are pasted on the seat of a wheelchair. All these sensors measure the pressure distribution continuously and give data on the pressure exerted by different parts of the body, which is very important for assessing the user’s posture.Ultrasonic Sensor: A unidirectional sensor is packed on the wheelchair’s backrest to measure the distance from the backrest to the user’s body. The sensor helps detect improper leaning related to some posture problems.Power Supply: The system’s 5 V-DC power supply or battery pack is an integral part, ensuring the continuous and reliable operation of the microcontroller and sensors during the monitoring process.Software specificationsArduino IDE: This integrated development environment codes the firmware and loads the code on the Arduino Uno microcontroller hardware. The code deals with the real-time reading of sensor values, the preprocessing of this data, and Wi-Fi module communication.ESP8266 Library: The ESP8266 library provides a layer of abstraction over programming the ESP8266 Wi-Fi module. It is necessary for sending acquired sensor data wirelessly over Wi-Fi to the cloud-based posture-monitoring system.The data collected, after post-processing, the EPO makes FS, and thereby, postures using the DMN are predicted, with the help of customized scripts in Python and MATLAB, towards data processing and analysis.Results and discussionsIn this study, the proposed EPIC framework has an experimental testbed with specially developed hardware and software that monitors the posture and health of paralyzed patients.Specifications of the simulation environmentThe developed data collection, processing, and analysis workflows are simulated in MATLAB and Python. The environment in these tools is controlled, allowing one to test the system’s actual functionality and ensure its accuracy using the collected data.The proposed work has been implemented at a sampling rate of 50 Hz; it is quite a normal sampling rate, reflecting a balance between the accuracy of data obtained and computational efficiency. Gaussian noise(5-10%) is added to the readings to test the robustness of the posture detection algorithm under less hostile than desired conditions in sensor readings.The Wi-Fi module’s data transmission latency of about 30ms is simulated to ensure the system is not delayed. The performance of the DMN model was better when compared to the baseline models on evaluation. DMN scored 94.3% accuracy, while the Baseline Model-1 scored 89.2%, and the Baseline Model-2 scored 85.7% in accuracy. The DMN scored 93.7% in precision, while Baseline Model-1 and Model-2 scored 88.4% and 84.8%, respectively. The mean recall was also higher for the DMN, at 94.5%, compared to 89.6% for Baseline Model-1 and 86.0% for Baseline Model-2. The corresponding F1-score that balances recall and precision was 94.1% for DMN, compared to 88.9% for Baseline Model-1 and 85.4% for Baseline Model-2.Also notable were the training and inference times. The DMN took about 6 h to train, slightly higher than Baseline Model 1 by 5.2 h and marginally shorter than Baseline Model 2 by 6.5 h. The inference time taken by the DMN was 45 ms, multiple times faster than the baseline models, which had inference times of 70 ms and 65 ms, respectively.The superior performance metrics of the DMN, as illustrated within the comparative graphs, reveal the network’s posture prediction functionality. The robustness of the DMN in authenticating the real patient posture from the sensor data is demonstrated by higher accuracy, precision, and recall rate. A balanced performance of the model and improved F1 score should be a reliable tool in the real-time monitoring of postures. Table 2 shows the test accuracy of DMN for various learning rates and dropouts.Table 2 Test accuracy of the DMN for different combinations of learning rates and dropout rates.Full size tableThe test accuracy for the DMN model differs significantly with different combinations of learning and dropout rates. The test accuracy of a learning rate of 0.01 with a dropout rate of 0.3 is 94.3%, just like a learning rate of 0.001 with a dropout rate of 0.4. These combinations indicate that a moderate dropout level is optimal to prevent overfitting while maintaining good performance of the model. Higher learning rates, say 0.01, generally show promising results, though the high dropout rate may cause slight overfitting. Conversely, lower learning rates, say 0.001, provide much stability and produce relatively consistent accuracy across different dropout rates. Specifically, the dropout rates 0.3 or so appear to hold special prominence in balancing regularization and the model’s learning ability. On the other hand, for dropout rates higher than 0.4, there is a slight drop in accuracy, probably due to the reduced learning from data by the model. All this again brings out an essential lesson on fine-tuning both the learning rate and dropout for optimal performance of the DMN.The efficiency of the DMN can be seen through its training and inference times. Although it requires slightly longer training time, fast inference time adds to the applications’ responsiveness in real-time decision-making. This advantage makes this model particularly suitable for application in the EPIC framework, where timely posture adjustments are essential in ensuring patient comfort and quality of care. The DMN showed the highest accuracy in the prediction of postures compared to the baseline models. High precision of the DMN indicates that there are fewer False Positives (FP) compared to the baseline models. A higher recall of the DMN demonstrates that it is better at detecting true positives. The F1-score of the DMN shows that this model performs well-balanced, considering precision and recall. While a bit longer than SVM, the training time for the DMN was comparable to the MLP model. DMN provides faster inference, which is advantageous concerning real-time applications, as shown in Fig. 6.Fig. 6Performance metrics comparison with SVM, MLP, and the proposed work.Full size imageTable 3 explicitly compares the DMN with other models in terms of accuracy, precision, recall, and inference speed.Table 3 Performance metric of the proposed work.Full size tableThis can be a simple classification algorithm, such as a Support Vector Machine (SVM) with a linear kernel. Another way is traditional neural network models with standard activation functions—for instance, ReLU—and fewer layers, like a multilayer perception. These are the simplest neural network models used to provide a baseline against which to compare more sophisticated architectures like DMN.Figure 7a,b show the proposed model’s training and testing performance plots on loss and accuracy metrics for 50 epochs. Both train and test losses lie high, which is above 2.0, and both of them drop significantly within the first 20 epochs. At about the 20th epoch, these loss values are deficient at ~ 0.2. The training loss decreases slightly faster than the test loss, implying that the model quickly picks up the learning trend from the training data. Again, both losses start converging and remain steady after the epoch at the 20 mark, establishing that the model has already created the zone of minimal improvement in loss.The graph starts at about 50% accuracy for the training and test sets. The training and test data trends increase drastically to over 90% within the first 15 epochs. At about the 20th epoch, the training and test accuracy values reach almost perfect levels of ~ 100% and remain stable, thus establishing that the model does excellently on both the training and test datasets. The model converges within 20 epochs both for loss reduction and accuracy improvement. After epoch 20, performance metrics stabilize; thus, the model converges to its optimal performance. Near-perfect accuracy and low loss values of the model over the training and test datasets may hint at well-tuned hyperparameters and probably good generalization. These numbers, therefore, correspond to a well-performing model in terms of accuracy and loss; the model can thus be said to learn and generalize from data effectively.Fig. 7Proposed model (a) accuracy (b) loss.Full size imageThe performance of a posture prediction classification model is depicted using the confusion matrix in Figure 8, where the true labels are on the Y-axis and the predicted ones are on the X-axis. There are three classes: “Upright,” “Leaning Forward,” and “Leaning Sideways.” Each cell contains the percentage of the model’s predictions to some specific true label.Perfect accuracy for the model was achieved in predicting “Upright” posture and “Leaning Forward,” with 100% of those cases correctly classified. In the “Leaning Sideways” posture, the model got 85.71% of the cases right, while 14.29% of the “Leaning Sideways” instances were predicted as belonging to “Leaning Forward.” This misclassification may imply that sometimes the model misidentifies these two postures, probably due to similarities in data features used for prediction. The model has high overall accuracy but may require some fine-tuning to trim errors while implementing the “Sideways Lean” and “Forward Lean” postures.Fig. 8Confusion matrix of the proposed work.Full size imageThe DMN results generally showed that it could efficiently utilize the FSRs and ultrasonic sensor data. The feature selection process of EPO ensures that only the most appropriate fraction of the data is supplied to the model. The results also recommend that embedding DMN within the EPIC framework can support very efficient posture monitoring and intervention among wheelchair-bound subjects. Future work may then use a more extended dataset in diverse postures and conditions, which could further improve the accuracy and robustness of the DMN. Research into other feature selection techniques and the integration of more sensors would further enhance posture and health monitoring understanding. These results have shown, through comparative graphs, the potential of a DMN to become a powerful tool in real-time posture prediction and monitoring systems.We have assessed the effectiveness of various machine learning models, such as, RF, and DNNs, in forecasting fatigue-related postural deviations. Employing confusion matrices, precision-recall metrics, and ROC curves allows for quantitative evaluation of each algorithm’s performance. Additionally, ANOVA tests investigated the statistical importance of diverse postural parameters at different fatigue levels. These statistical studies indicate that AI-based posture monitors are good at predicting what will happen and how useful they are for finding early signs of fatigue and determining the risk of ergonomic problems, especially at work.Comparative discussion with existing studiesRecent advancements in assistive healthcare technologies have utilized diverse techniques for recognizing posture in wheelchair users, including Graphical User Interface (GUI)-based feedback systems, independent component analysis with knowledge distillation, and fundamental sensor-fusion models. For instance, GUI-based systems frequently depend on visual analysis and manual alert processes, restricting their scalability and efficiency26. On the other hand, our EPIC system provides automated real-time detection and alerting through an Android-integrated IoT platform, allowing caregivers to monitor remotely and continuously.Table 4 Analysis of the proposed EPIC framework with existing posture monitoring systems.Full size tableAlthough the independent component analysis technique is proficient at condensing feature representations for training purposes25, it falls short in adapting to select the best features from varied multi-sensor data inputs. Our method employs the EPOA for dynamic feature selection, improving computational efficiency and classification outcomes. Additionally, our model’s deep Maxout network enhances prediction accuracy by surpassing conventional activation functions when dealing with nonlinear sensor variations20.The proposed EPIC framework exhibits a 10.1% increase in accuracy compared to traditional posture recognition techniques. Additionally, it surpasses GUI-based monitoring systems26 by 7.73% and independent component analysis strategies by 2.84%, as demonstrated by our evaluation metrics. These enhancements are due to the smooth integration of optimized sensor data with deep learning, establishing EPIC as a more dependable solution for real-time health monitoring.This comparative study as shown in Table 4, demonstrates that the EPIC system enhances classification metrics while providing a stronger, more scalable, and intelligent solution for posture monitoring. Our approach’s modular nature allows it to be adapted for other healthcare sectors, including rehabilitation monitoring, detecting falls in the elderly, and assessing ergonomics in workplace settings. Therefore, this methodology is promising for replication and growth across diverse applications, aiding future R&D in smart assistive technologies.Conclusion and future workIn this study, the proposed framework of the Emperor Penguin optimized sensor Infused Wheelchair (EPIC) has an experimental testbed with a specially developed hardware and software setup that works as a monitoring tool for the posture and health of paralyzed patients. Also integrated with the framework for feature selection is the Emperor Penguin Optimizer Algorithm (EPOA). Regarding test accuracy, the Deep Maxout Network (DMN) model performed very well in the posture prediction task in wheelchair-bound subjects. This is a considerable improvement over the baseline models, wherein Baseline Model-1 stood at 89.2%, and Baseline Model-2 stood at an accuracy of 85.7%. Moreover, the DMN model showed a high precision of 93.7%, a high recall of 94.5%, and a high F1-score of 94.1%, underpinning the model’s strength in classifying the correct postures. Despite these strong results, the confusion matrix analysis has identified a minor challenge in the “Leaning Forward” versus “Leaning Sideways” postures, where 14.29% of instances of the “Leaning Sideways” category were misclassified. This demonstrates that although the DMN performs very well, there is room for improvement in handling the same posture classes.In the future, more diverse data sources, new feature selection methods, and multiple variants of deep learning architectures will be used to deal with the challenge. This will further increase the model’s accuracy in differentiating subtly different postures, making it even more reliable for real-world applications.Data availabilityAll data generated or analyzed during this study are included in this published article.ReferencesMiquel, M. V. Advancing Coastal Inundation Forecasting: A Multifaceted Machine Learning Approach. PhD Diss (Texas A&M University-Corpus Christi, 2024).Bilal, A. et al. El-Sherbeeny, and Haixia long. BC-QNet: A quantum-infused ELM model for breast cancer diagnosis. Comput. Biol. Med. 175, 108483 (2024).Article PubMed Google Scholar Sharma, R. & Chander Andreia de bem machado, Isabel Sofia brito, Fernanda Maria Dos Santos pereira, and Maria José sousa. Knowledge representation on innovations in learning processes and educational policies: infusing artificial intelligence and blockchain powers. In Incorporating AI Technology in the Service Sector, 45–66. Apple Academic, (2024).Kannan, K. D. et al. Advancements in machine learning and deep learning for early detection and management of mental health disorder. ArXiv Preprint arXiv:2412.06147 (2024).Zhao, T., Faustino, B. L., Jagatheesaperumal, S. K., Rolim, F. P. S. & de Albuquerque, V. H. C. Reinforcement learning-based adaptive deep brain stimulation computational model for the treatment of tremor in parkinson’s disease. Expert Syst. Appl. 267, 126154 (2025).Afridi, S., Hlebowicz, K., Cawthorne, D. & Ulrik Pagh Schultz Lundquist. and. Unveiling the Impact of Drone Noise on Wildlife: A Crucial Research Imperative. In International Conference on Unmanned Aircraft Systems (ICUAS) 2024. (2024).Huang, J. et al. A medical decision support platform for identifying thrombospondin 1 in non-alcoholic fatty liver disease via immuno-infiltration analysis. J. Vis. Exp. (JoVE) 210, e67328 (2024).Jagatheesaperumal, S. K., Rahouti, M., Ahmad, K., Al-Fuqaha, A. & Guizani, M. The duo of artificial intelligence and big data for industry 4.0: Applications, techniques, challenges, and future research directions.IEEE Internet Things J. 9(15), 12861–12885 (2021).Vu, T. H. et al. Applications of generative AI (GAI) for mobile and wireless networking: A survey. IEEE Internet Things J. 12(2), 1266–1290 (2025).Ganesan, P. et al. Deep learning-based interactive dashboard for enhancing online classroom experience through student emotion analysis. IEEE Access. 12, 91140–91153 (2024).Article Google Scholar Priyanka, G., Kumar, J. S. & Veena, S. T. Deep learning based video surveillance for predicting vehicle density in real time scenario. J. Ambient Intell. Humaniz. Comput. 14(4), 4371–4383 (2023).Sunil, L. & Tade., Vibha, V. Hybrid deep emperor Penguin classifier algorithm-based image quality assessment for visualization application in HDR environments. IET Image Proc. https://doi.org/10.1049/IET-IPR.2019.1371 (2020).Article Google Scholar Ganesan, P., Jagatheesaperumal, S. K., Gaftandzhieva, S. & Doneva, R. Novel cognitive assisted adaptive frame selection for continuous sign language recognition in videos using ConvLSTM. Int. J. Adv. Comput. Sci. Appl. 15(7), 418–425 (2024).Sofia, J., Priya, D. & Subramanyam, M. V. Emperor Penguin optimized user association scheme for MMWAVE wireless communication. Wireless Pers. Commun. https://doi.org/10.1007/S11277-020-07269-3 (2020).Article Google Scholar Milos, D., Dusan, K., Maria, S., Nuno, M. & Garcia Real-time wireless UWB sensor network for person monitoring. https://doi.org/10.23919/CONTEL.2017.8000034 (2017).Kingsly, A., Anuradha, J., Suganthi, B. B. & Balasaranya, K. Modified Penguin search algorithm to improve feature selection in wireless sensor network. (2023). https://doi.org/10.1109/icecaa58104.2023.10212419.Priyadharsini, S., Jagatheesaperumal, S. K., Alkhamees, B. F. & Hassan, M. M. Foreign object debris detection in lane images using deep learning methodology. PeerJ Comput. Sci., 11, Art. no. e2570 (2025).Calpakkam, S. R., Kanagaraj, Y., Raja, P. A. & Jayasankar, B. T. An optimized deep neural network-based DoS attack detection in wireless video sensor network. J. Ambient Intell. Humaniz. Comput. https://doi.org/10.1007/S12652-020-02763-9 (2021).Article Google Scholar Hwajeong, S. & Howon, K. Optimized implementation of HIGHT algorithm for sensor network. J. Korean Inst. Inform. Communication Eng. https://doi.org/10.6109/JKIICE.2011.15.7.1510 (2011).Article Google Scholar Syed, H., Gaurav, S. & Harshita, K. Optimized Sensor Network Design for Real-Time Environmental Monitoring. https://doi.org/10.1109/icocwc60930.2024.10470648 (2024).da Silva, D. S., Nascimento, C. S., Jagatheesaperumal, S. K. & de Albuquerque, V. H. C. Mammogram image enhancement techniques for online breast cancer detection and diagnosis. Sensors, 22, 22, Art. 8818 (2022).Saif, W., Ali, A. & Adel, G. Enhancing intrusion detection with LSTM recurrent neural network optimized by emperor Penguin algorithm. Wasit J. Comput. Math. Sci. https://doi.org/10.31185/wjcms.166 (2023).Article Google Scholar Satinder, Singh, M., Sonia, G. & Ranjit, K. Optimized sensor nodes deployment in wireless sensor network using Bat algorithm. Wireless Pers. Commun. https://doi.org/10.1007/S11277-020-07823-Z (2021).Article Google Scholar Boje, E., Hardy, S. & Jason, A. A Sensor and a Sensor Network for Monitoring Utility Poles. (2019).Ze, Z., Arlington, M., Divya, F., Geraint, T. & Ashutosh, J. Tiwari. Sensorformer: A Memory-efficient Transformer for Industrial Sensor Fusion. https://doi.org/10.1109/ICAC55051.2022.9911157 (2022).Hansa, S., Hansa, S., Hetal, P. & Hetal, P. Emperor Penguin optimized event recognition and summarization for cricket highlight generation: multimedia systems. https://doi.org/10.1007/S00530-020-00684-3 (2020).Ryzin, P., Van Patrick, A. & Van, A. Ryzin. (2003). Medical monitoring network sensor interface device.Mason, P. Clear Bright Future: A Radical Defence of the Human Being (Penguin UK, 2019).Dhiman, G., Vinoth Kumar, V., Kaur, A. & Sharma, A. DON: deep learning and Optimization-Based framework for detection of novel coronavirus disease using X-ray images. Interdiscip Sci. 13 (2), 260–272. https://doi.org/10.1007/s12539-021-00418-7 (2021). Epub 2021 Feb 15. PMID: 33587262; PMCID: PMC7882874.Article CAS PubMed PubMed Central Google Scholar Sangeethaa, S. N. Presumptive discerning of the severity level of glaucoma through clinical fundus images using hybrid PolyNet. Biomed. Signal Process. Control. 81, 104347. https://doi.org/10.1016/j.bspc.2022.104347 (2023). ISSN 1746–8094.Article Google Scholar González-Cely, A., Ximena, Camilo, A. R. & Diaz Mauro Callejas-Cuervo, and Teodiano Bastos-Filho. Optical fiber sensors for posture monitoring, ulcer detection and control in a wheelchair: a state-of-the-art. Disabil. Rehabilitation: Assist. Technol. 19 (4), 1773–1790 (2024).Google Scholar Hou, L. et al. Auton. Wheelchair Health Monit. Syst. Based Internet Things Sci. Rep. 14, 1 : 5878. (2024).MathSciNet CAS Google Scholar Bhaskar, N. et al. Real-Time Remote-Controlled Wheelchair with Wi-Fi Integration and Obstacle Detection for the Disabled and Elderly People. In 2024 2nd International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT), pp. 321–325. IEEE, (2024).Vermander, P., Mancisidor, A., Cabanes, I. & Perez, N. Intelligent systems for sitting posture monitoring and anomaly detection: an overview. J. Neuroeng. Rehabil. 21 (1), 28 (2024).Article PubMed PubMed Central Google Scholar Jabade, V., Shrivastav, H., Rupani, J. & Harak, N. Mobility Plus: Voice-Controlled Wheelchair with Health Monitoring System and Oxygen Cylinder Integration. In 2024 International Conference on Inventive Computation Technologies (ICICT), pp. 2167–2171. IEEE, (2024).Chaitanya, N. et al. IoT-enabled Moving Wheelchair with Obstacle Detection and Continuous Health Monitoring. In Second International Conference on Emerging Trends in Information Technology and Engineering (ICETITE), pp. 1–7. IEEE, 2024., pp. 1–7. IEEE, 2024. (2024).Download referencesAuthor informationAuthors and AffiliationsDepartment of ECE, Indian Institute of Information Technology - Design and Manufacturing (IIITDM), Jagannathagattu Hill, Kurnool, 518007, Andhra pradesh, IndiaVijaya GunturuDepartment of Computer Science & Engineering, Koneru Lakshmaiah Education Foundation, Bowrampet, Hyderabad, 500043, Telangana, IndiaJ. KavithaDepartment of Electronics and Communication Engineering, CVR College of Engineering, Hyderabad, Telangana, IndiaSwapna ThoutiDepartment of Computer Science & Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology, Chennai, 600 062, IndiaN. K. Senthil KumarDepartment of College of Science and Engineering, Southern Arkansas University, Magnolia, AR, 71753, USAKamal PoonComputer and Network Engineering Department, College of Computing, Umm Al-Qura University, Makkah, 21955, Saudi ArabiaAyman A. Alharbi & Amar Y. JaffarDepartment of Computer Science, Dambi Dollo University, Dambi Dollo, EthiopiaV. SaravananAuthorsVijaya GunturuView author publicationsYou can also search for this author inPubMed Google ScholarJ. KavithaView author publicationsYou can also search for this author inPubMed Google ScholarSwapna ThoutiView author publicationsYou can also search for this author inPubMed Google ScholarN. K. Senthil KumarView author publicationsYou can also search for this author inPubMed Google ScholarKamal PoonView author publicationsYou can also search for this author inPubMed Google ScholarAyman A. AlharbiView author publicationsYou can also search for this author inPubMed Google ScholarAmar Y. JaffarView author publicationsYou can also search for this author inPubMed Google ScholarV. SaravananView author publicationsYou can also search for this author inPubMed Google ScholarContributions“Conceptualization, V.G, J.K and N.K.S; methodology, K.P software, A.A.A.; validation, A.Y.J, and A.A.A; formal analysis, V.G; investigation, J.K; resources, A.A.A; data curation, S.T; writing-original draft preparation, V.S; writing—review and editing, V.S; visualization, K.P; supervision, N.K.S; project administration, S.T; funding acquisition, A.Y.J. All authors have read and agreed to the published version of the manuscript.”Corresponding authorCorrespondence to V. Saravanan.Ethics declarationsCompeting interestsThe authors declare no competing interests.Consent for publicationInformed consent from all subjects for publication of identifying images in an online open-access publication.Human and animal rightsNo animals/humans were used for studies.Additional informationPublisher’s noteSpringer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Rights and permissionsOpen Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.Reprints and permissionsAbout this article