DOI:10.32604/csse.2022.022234
| Computer Systems Science & Engineering DOI:10.32604/csse.2022.022234 | |
| Article |
Virtual Nursing Using Deep Belief Networks for Elderly People (DBN-EP)
1Department of Information Technology, KGiSL Institute of Technology, Coimbatore, 641035, India
2Department of Computer Science and Engineering, Coimbatore Institute of Technology, Coimbatore, 641014, India
*Corresponding Author: S. Rajasekaran. Email: proffraja@gmail.com
Received: 31 July 2021; Accepted: 30 September 2021
Abstract: The demand for better health services has resulted in the advancement of remote monitoring health, i.e., virtual nursing systems, to watch and support the elderly with innovative concepts such as being patient-centric, easier to use, and having smarter interactions and more accurate conclusions. While virtual nursing services attempt to provide consumers and medical practitioners with continuous medical and health monitoring services, access to allied healthcare experts such as nurses remains a challenge. In this research, we present Virtual Nursing Using Deep Belief Networks for Elderly People (DBN-EP), a new framework that provides a virtual nurse agent deployed on a senior citizen’s home, workplace, or care centre to help manage their health condition on a continuous basis. Using this method, healthcare providers can assign various jobs to nurses by utilizing a general task definition mechanism, in which a task is defined as a combination of medical workflow, operational guidelines, and data gathered from a remotely monitored virtual nursing system. Practitioners are in charge of DBN-EP and make treatment decisions for patients. This allows a DBN-EP to act as a personalized full-time nurse for a client by carrying out practitioner support activities based on information gathered about the client’s health. An electronic Personal Health Record (ePHR) system, such as a specialized web portal and mobile apps, could provide such patient information to elderly person family members and care centres. We created a prototype system using a DBN-EP system that allows traditional client applications and healthcare provider systems to collaborate. Finally, we demonstrate how this system may benefit the elderly through a result and debate.
Keywords: Deep belief networks; RBM; video mining; elder people; elder care
The universal care method has become more complex, with increased operationalization and treatment quality [1]. Before employing the technology, patients rely on their traditional senses of vision, feeling, and hearing to monitor their health and movements. Changes in the patient’s physical health have heightened the powerlessness of care providers over time [2]. Nurses will use tiny mental or skin changes to detect early blood oxygen differences when preparing for admission. Pulse oximetry really assists personnel in tracking decreased oxygenation before clinical indications appear, allowing for faster root identification and processing [3].
According to F&S researchers, health insurers use Big Data Analytics and other analytical solutions to extract operational knowledge from these data. Patient monitoring has progressed from ad hoc to continuous multiple criterion monitoring, allowing clinicians to increase the amount of unprocessed and unorganized data they collect. According to the report, by 2020, health systems will invest over million in technologies that allow such analytics to both assess patients in the moment and hint at a possible future diagnosis [4].
As more elderly people live alone, elder care at home is becoming a major issue, as accidental injuries may occur, putting their health at risk. To expand therapy in an accessible and cost-effective manner, it is critical to develop technologies that assist elderly people individually. In geriatric therapy applications, the use of an event-driven technique to track ambient and patient activities in real time is common [5]. It is critical to analyse the methodologies of the aged care literature field in a new area of study and development in order to describe existing strategies for future research routes. This planned investigation would also include a detailed survey of wearable sensors for various senior care programmes. This strategy is an attempt to obtain insight into several types of home monitoring systems for older persons that rely on environmental sensors. For treatment systems, ageing in place is a viable alternative. According to the Centers for Disease Control, it is a privilege to live joyfully, financially, and comfortably with one’s own house and community, regardless of age, money, or talents [6]. Due to emotional concerns, a big number of older people have chosen to grow old and keep their independence in their own homes, which has sparked a lot of interest in the topic of ageing in recent years. Elderly social well-being and carefully understood social media are encouraged as they age, so that they do not lose their living conditions in a familiar environment. The new policy prioritises environmentally friendly living facilities (AAL), which have seen significant improvements in recent years. AAL technologies monitor and assist with daily living activities (ADL) to prevent, cure, and improve the well-being and health of the elderly [7].
The research project is finished, and the findings will be put into action. As part of the research, the majority of the programmes used to monitor senior accidents (for example, behavioural lapses in everyday life) were documented for the development of independent living. Non-contact sensor technologies include motion, pressure, video, contact object, and sound sensors, to name a few. Smart technologies and multi-component techniques (e.g., wearable sensor combinations) have also been developed. Sensors, in addition to room-assembled ambient sensors, are reported by robot-based senior care tasks. Research on the use of older behavioural monitoring systems is widespread, but it is still in its infancy and primarily consists of small-scale experiments [8]. Another exciting market is elderly behaviour monitoring technology, particularly for long-term elderly care. Monitoring technologies with more extensive analyses that analyse and demonstrate their potential to contribute to the extension of older people’s autonomous lives, on the other hand, should be taken to the next level [9]. Smart watches with embedded nanosensors, such as automated glucose monitoring, heart rate monitors, infusion pumps, and ECG monitors, are a good place to start for health care providers to manage contagious diseases and enhance health both inside and outside the hospital [10]. New technologies like intelligent prosthesis and intelligent implants, on the other hand, are gaining popularity. These are vital to patient therapy after surgery or rehabilitation by assisting in the measurement of essential requirements to permit follow-up and early action, reducing readback or problems. Wireless tablets and nano mobiles [11] are another invention that has been produced to provide a rapid rise in the detection of drug permeability. According to the paper, new difficulties in cardiac and emergency care are being posed by innovative technology and intelligent materials. Brain-computer interface implementations can explicitly monitor and measure core health measures to assess a patient’s physical, psychological, and mental status. The nicest thing about systemic risks is remote monitoring and being out of patient rooms. Staff is, without a doubt, the most expensive aspect of healthcare. You have a lot less employees if you don’t have hospital wards. Set up video tracking surveillance centres-with the right equipment, this can be spectacular and quite inexpensive [12]. The provincial administration also recognizes the importance of patient monitoring. The expansion of remote healthcare services, according to the researchers, is necessary to ensure that domestic health facilities employ their innovation to give state-of-the-art care. When almost any lifetime function is watched and the data obtained and saved, patient data monitoring will be merged with rival sources from a number of other devices in the future.” Data bursts will be manipulated and used by technologies such as Artificial Intelligence (AI), deep learning, and others to give individualized, guided therapy [13].
MHealth is predicted to be used by much more than half of the voters in developing nations by 2025, as fitness trackers, virtual wellness, social networking platforms, and patients gain traction. Patient surveillance is expected to outperform many organisations by 2025 [14]. This ensures billions because app sales are more likely than commodity sales. However, the drop in data is just the beginning of what healthcare professionals and clinics will have to deal with in the near future if they want to fully profit from the rapidly evolving patient monitoring technologies. Software Aids Diagnostics (CAD) has been a hot topic in image processing, clinical obstetrics, and other medical specialities, including dermatology, over the past decade [15]. The major purpose of CAD is to give machine output for the clinician to aid in image interpretation by improving clinical radioactivity outcomes and clarification while also lowering the amount of time necessary to view the picture. Because CAD may be applied to all imaging methods, locations, and tests this fascinating feature of digital image processing will likely have a large impact on medical imaging and diagnostics in the next decades [16]. This paper is organized as follows: Section II explains the Related Work of nursing to elders and techniques regarding Deep Belief Networks. In Section III, Research Methodology Discuss about the video analytics of the elders using Deep Continuous Deep Belief Network (DC-DBN) with Restricted Boltzmann Machines (RBM). Using these techniques the various scales of Elderly person movements and video motions are captured from various camera’s are analysed and the features are extracted [17]. Experimental Results and Discussion are shown in Section IV, where the contributions about the proposed work are done. The methodology and equations utilized in the proposed work and the results were shown in the tabular format where the person’s activities are monitored using live camera and behaviour are analysed and finally in Section IV conclusion done.
According to studies on the ageing population, the percentage of elderly people without a caregiver has increased [18]. These patients are nevertheless at a significant risk for unfavourable events such as high blood pressure, various stroke-related health issues, and other body instability injuries, which can lead to dangerous falling. If an unmanaged decline is not reported to a care facility right away, it might lead to even worse conditions than the original reason. Over the last ten years, there has been a significant increase in demand for smart systems to prevent, track, and monitor accidents in order to reduce the negative consequences of these unfortunate incidents.
Several researches have been proposed so far that address various parts of the topic of fall detection, ranging from simple systems to advanced detection algorithms and extraction approaches. With this article, a mechanism is created to intelligently detect, classify, and track elderly slips. After pre-processing and scheduling the data with a predetermined window, the triaxial intensity of the human motions is analyzed using a private mobile, and the relevant attributes are generated. Next, the profound belief network (DBN) uses two public data sources, nine falls, and one activity level class for training and validation. In comparison to the other 9 research referenced, simulation findings from TFall and MobiFall two standardized datasets have a sensitivity level of 97.56 percent and a precision level of 97.03 percent, which is favorable.
Recent advances in vision sensor technology have enabled Mohamed Eldib’s work on Comportment research for home surveillance applications. The present generation of home-based surveillance equipment has some flaws. Digital sensor calibration is a time-consuming technique that is impractical in everyday life, resulting in recalibration when a caregiver or old person inadvertently changes visual sensors, safety issues, and high hardware installation costs. For long-term behavioural testing, they recommend employing a low-resolution system with low-cost sensor sets. Starting with a visible individual attachment to the front/background sensor that registers the movement level of each object, action analyses begin. After then, the hidden Markov (HMM) prototype will be utilised to calculate the consumer’s position. Finally, [19] suggests a method for conducting experiments in frequency and spectrum contexts. For the past ten months, we’ve been conducting real-world tests. They show that the HMM solution works for 30 days against earthly existence and the designation of the closest neighbour. Every other system distinguishes 13 daily activities (ADL variables). The versatile trends and certain primary ADL characteristics in order to diagnose rising or worsening health concerns.
The Visual, Auditory, Infrared, and heartbeat-based Old Monitoring Device developed by Dongwei Lu is based on multi-information fusion technology to address children’s lack of time and resources in caring for the elderly. In conjunction with speech recognition, infrared recognition, and pulse detection results, video processing technology was utilised to track old people’s life in real time. In the event of an emergency, the device will handle the difficult task of sending data to the server and communicating with youngsters via GPRS (General Packet Radio Service). The computer creates the Software mobile in order to scan the old man’s life in real time [20].
According to Kim’s results, as the world’s population ages, an increasing number of older adults—who have lower insulin sensitivity and a poorer sense of balance than teenagers—are participating in social activities and recovering from fall injuries. As a result, they both died as a result of their injuries. Fracture is one of the four leading causes of death in the elderly, and even if it does not result in death, it can result in fatal accidents, thus the system must be able to identify falls that prevent bone fractures. As a result, a method for detecting falls has been developed in order to properly measure and track falls [21]. Three-axis elevation detectors and six-axis tilting sensors were employed in the proposed fall proximity sensor, which measured body motions caused by falls and identified falls using falling detection techniques.
In the fall tests, Zhou Li-findings peng’s were generated utilising a fall induction device based on a plane-pressure actuator, which induced the participants’ natural fall. The rotational speeds and impulses of the several essential sections of the body produced during the fall have been measured, analysed, and integrated in order to measure the falls and their movements [22]. To distinguish fall from normal life practises, the momentum and inclinations produced during daily lifespan activities were computed and compared to those produced during falls to develop fall detection algorithms. With specific techniques, the fall detection method has been able to distinguish falling from regular behaviour such as walking, riding, sitting, standing up, and lying. The machine was able to track the course of the fall. The method is anticipated to be useful in the application of our proposed framework for avoiding broken bones.
Malu, The identification of fallen individuals has become a crucial concern in the medical and health industries due to the exponential development of the ageing population. Recent detection of deteriorating events in video surveillance and in-time feedback would greatly reduce accidents, especially fatalities caused by old people. An enhanced FSSD (Feature Single Shot Multibox Detector) fall detection system was proposed [23] in light of the numerous circumstances of video surveillance and varied human conducts associated with this research. For starters, the data gathering is predicated on the generation of a video frame using a different dropping video sequence. The sample set is applied to the enhanced neural network even before the network converges. Finally, the target community and endpoint in the video are evaluated using an integrated network model. The results of the experiments show that the improved FSSD approach can detect the drop-off or ADL of picture frames as well as provide real time recommendations. The detector frequency (GTX1050Ti) is 24 frames per second, allowing it to meet real-time requirements while retaining measurement precision. In comparison to the complex FSSD techniques, the result is better than other techniques. The ability to forecast failure acts in video further validates the deep learning process’ effectiveness and reliability [24].
Classify Video Content Using Deep Belief Networks to Care Elders
In truth, nothing matches our brains’ production of real-time data and rhythms. Designers are currently uncovering patterns that replicate our subconscious network, which are comprehended by fundamental deep learning in their scientific progress. Deep learning models are distinguished by the presence of a comparatively extensive and complicated portion between the hidden and output layers. This secret piece must be contained in a deep convolutional neural network with at least two layers. Any relationship between nodes and nodes specifies the strength of the interaction between the different nodes. The confidence interval for all connections to a single node is defined and converted to a number between zero and one, which is then used to create an activation function. The output is subsequently transferred to a different network. This process has progressed to the growth nodes. The nodes of growth are activities that elders notice, such as rapid collapse, disease, or a deep attack. As the model recognises, the weight between the linkages is always changing. Because of their composition, deep neural networks are more capable of spotting patterns than shallow networks, as illustrated in Fig. 1. Deep neural networks recognise information-based data after being trained on named data. To put it another way, they may learn from their mistakes rather than needing to create special guidelines for each task. For example, if the model is conditioned for an elderly person’s role behaviour, the model will learn to recognise the general aspects of the seniors, such as the pointing body, general form, and limb movement, by training the model by adding abrupt collapse, heart illness on the mark.
Figure 1: Deep continuous deep belief network (DC-DBN) with RBM model for video content mining
Although most deep neural networks are unidirectional, recurrent neural networks allow information to flow in any direction. Knowledge is mirrored in these networks via their memory, therefore earlier behaviour has an impact. This qualifies them for tasks such as word recognition and handwriting. Because of the senior’s in-depth system study, RNN loses its perception of understanding and effectiveness. CNNs are models of the visual cortex in the human brain that are frequently employed during visual experiences. When consumers look at a frame, they will break it down into little bits to recognise and distinguish the image’s important components. CNNs compress the size of an image so that it can be examined more quickly without losing important details. Within a convolutional neural network, the secret layers are referred to as convolutionary layers, which increase the filtration capacity of the individual layers. The basic patterns are described by the first convoluted layers, while later designs mix them. Deep belief networks are designs that use uncertainties to produce outputs and learn unsupervised. They are made up of endogenous attribute values, which can be found in both unfired and directed layers. Unlike conventional models, deep networks learn the entire data from each layer. The earliest structures in a convolutional deep neural network simply choose streams for unique properties such as borders, and the contributes to a better reassembling of all the common qualities detected in prior layers. Simple belief networks, on the other hand, regulate each layer globally. The node is similar to a set of Boltzmann Minimal Equipment, in which networks are linked to all previous entities and accompanying surfaces. In contrast to RBMs, nodes interact laterally within their layers’ deep network of confidence. Multiple layers are connected by a network of symmetrical weights. The relationships at the top are undifferentiated, and their relationship is formed by their relationship. The links are sent to the bottom of the page. With previous networks and translucent layers with eastern borders, nodes in the concealed layer and as an occult layer perform two tasks. The links in the results are identified by these nodes. Deep faith networks are pre-trained using greedy deep learning. This is an approach that solves problems by looking for the best option for each layer in the sequence and then finding the best overall answer. The course is implemented layer by layer, influencing the layers of the deeper belief networks one by one. The output from the preceding layer is used as input in each step, and each layer is supplied with a distinct version, including its data. Even though greedy deep learning is fast and accurate, it is employed to create simple beliefs. They also aid in the weight reduction of any frame.
Boltzmann machines (BMs) have a linear variant of the log-linear Markov Random Field as their functional form (MRF). We discover that most of the components are never visible (thus the name secret) and that by making them powerful enough to characterise complicated distributions (i.e., from a small parametric configuration to a non-parametric one), we may improve the Boltzmann machine’s modelling capacity (also called hidden units). Restricted Boltzmann Machines (RBM) also restrict BMs and those with both visible and hidden connections. Fig. 2. depicts a graphic representation of an RBM.
Figure 2: Restricted Boltzmann Machines (RBM) video image pixel content analysis
An RBM’s energy function E(v, h) is defined as:
That is.,
Here W reflects the linking weights of the concealed and exposed components and b, c are the coordinates of both exposed and invisible layers.
The corresponding free energy equation is incorporated directly into:
The obvious and secret groups are provisionally distinct from each other and regardless of their special arrangement. This asset is used,
We get from Eqs. (3) and (2) in the common study elders video frame analysis of dichotomous units, variations of the standard neuron activation mechanism probabilistically in Tab. 1:
For an RBM with differential units, the corresponding log-like grayscale images:
For image recognition, deep creed networks can be employed. An picture would be used as both the input and output form. From simple features like photo organisation to critical capabilities like medical diagnostics, this system has a wide range of applications. Smart microspores that can recognise images could be used to categorise viruses for Elder’s image frame analysis. This eliminates the need for specialised specialists in global epidemics, as well as the time it takes to respond. Deep belief networks are also used to identify falls in the elderly. Video recognition works similarly to vision in that it interprets video details as demonstrated in Fig. 3. by the DC-DBN. An object or a person’s gesture, for example, can be noticed. It has a wide range of applications, including home security, safety, and health care. Monitoring the orientation of ageing individual objects or the dropping identification of those objects, as well as the application of deep belief networks, are all examples of motion capture data. Motion capture is difficult because a computer can easily lose track of a person falling slowly while another person falls slowly owing to a similar sickness or if something is obstructing their vision suddenly. Thus, motion capture is dependent not only on the appearance of an elderly creature or human, but also on speed and direction. In the generation of picture object tracking and image analysis, motion capture is often used. RBM can be stacked and gullibly conditioned to generate so-called DBNs, according to the findings. DBNs are visual templates that learn how to interpret training data in a hierarchical manner. This is the mutual distribution between the vector x detected and the ℓ hidden layers hk:
Figure 3: Model of DC-DBN networks to care elders
where, x = h0, P(hk−1 | hk) is a dependent dispersion for observable units at RBM level k and P(hℓ−1, hℓ) is a hidden-visible projection at RBM level professional rate. The following figure shows this. DBNs with RBMs are the foundation of a greedy-layer non-regulated training principle for any layer. Learn the whole first surface of the input image x = h(0) as its feature map as an RBM. Using the first layer to display the information for use as data for the top level. There have been two common choices. The mean p(h(1) = 1 | h(0)) stimulation or p(h(1) = 1 | h(0)) activation can be chosen as a reflection. For training images (for both the visible RBM surface and the RBM surface), build the two layers as RBM, utilising converted data test results or standard detections. Each time upward, samples or mean values would be applied (Camera 4 and 5) for the necessary number of layers. Wrap up all variables for substituting a DBN log’s likelihood in a classification algorithm or approving a controlled testing criterion, as well as information to the rest machine for converting eligible descriptions into projections, as shown in Fig. 4. DBF is concerned with fine-tuning the regulated gradient of descent. We explicitly use the regression analysis classifier to describe the input x on the basis of the performance of the last hidden layer h^}(l)} of the DBN. By fine tuning the negative log-like cost function, a controlled gradient descent can then be achieved. Given that the supervised gradient is only non-zero for the weights and hidden units preconceptions of each layer that is null with relation to each visible RBM bias value, this strategy correlates to deep MLP setup with unregulated training method weights and hidden layer biases.
Figure 4: Elderly person movements at various scales analysis every specified duration of video motions captured from various camera
This can be seen that when all extracted features are initialised,
4 Experimental Results and Discussion
Begin by defining the DBN class and its associated RBMs, which will contour the MLP layers. Since we explore the perspective of using RBMs to initialise an MLP, the framework would demonstrate this by distinguishing as much as possible between the RBMs used to setup the system and the MLP used during grouping. A deep belief network is created by stacking numerous RBMs on top of each other. At layer X the activation function of the RBM then becomes origin of the RBM with ‘x + 1’ layer. The first RBM layer receives the Net’s feedback, and the hidden layer of the last RBM reflects the output. When each DBN is treated as an MLP by adding a logistic that is utilised for grouping, a layer of analysis is applied on top. Layers of self-sigmoid that form forward graphs such as those illustrated in Figs. 6 and 7. That collectively shape the MLP will be processed, whereas layers of self-RBM will be processed by the RBMs that used it to pre-process each framework of the MLP. Next stage, create n layers hyperbolic tangent layers use the recursive neural hidden Layer class, with a single adjustment that replaces the non-linearity of tanh with both the expeditionary feature
Figure 5: Various rooms person movements (hand, leg, eyesight…) with their behaviours (walking, sitting, sleeping,..) analysis by DC-DBN
Figure 6: Different camera live movements (hand, leg, eyesight..) of persons with their behaviours (walking, sitting, sleeping,..) analysis by DC-DBN
Figure 7: Camera 1 & 2 video movement and sound based classification person activity by DC-DBN
Self-definition, train set x, and affordable mobility variables, as well as k. pre-training functions Create a list of functions to perform a conjugate gradient step at a specific layer. The feature is to join the dataset’s index, and in order to train an RBM, the accompanying feature must be iterated Indexes in dataset. Currently, every pre-train function ns[x] feature accepts the indices as arguments and lr the hidden nodes optionally different camera detection shown in Fig. 6. Recognize that when the values are built, the details of the variables table (e.g., lr) are given, not the name of the python variables (e.g., learning rate), as shown in Figs. 7 and 8. If you have the amount of gerrard measures to be performed in a CD or PCD, k is also the claim for your feature. The describe build fine-tune functions claim self, databases, batch size, and learning rate. Continues to generate a practical train that goes through one phase of fine tuning and validates a feature that measures an error. Tab. 3 depicts the verification collection bunch and the feature evaluation calculates a mistake on a test sample collection. This framework, as illustrated in Figs. 9 and 10, is made up of two stages, the first of which is a category pre-work performance and the second of which is a customisation process. During the first training procedure, we loop through all network layers as shown in Fig. 11. Tab. 4 shows a comparison of the proposed DC-DBN with several existing techniques.
Figure 8: Camera 3 & 4 video movement and sound based classification person activity by DC-DBN
Figure 9: Camera 4 & 5 video movement and sound based classification person activity by DC-DBN
Figure 10: Camera 5 & 6 video movement and sound based classification person activity by DC-DBN
Figure 11: Detailed analysis of all camera live factors in the variousrooms movement and sound based classification person activity by DC-DBN
To define the RBM input at the x stage and conduct a CD-k step in this RBM, we employ the compile threshold d feature for any layer. For a certain set of newly-trained epoch times, this role is employed in the training set. The DBN category also includes a technique for generating the required functions (train model, validate model, and test model function). Throughout this phase, we use the internet to track a variety of patient criteria. The medical monitoring device, which is part of the IoT technology revolution, sends real-time health data to the cloud via the web. The criteria are transferred to a centralised database, allowing users to access data from anywhere in the universe. Between SMS-based medical monitoring and IoT patient monitoring systems, there is a significant difference.
Many patients can receive patient well-being information through the IOT framework. The reason for this is that a website or URL must keep track of the information. However, medical monitoring through GSM or SMS is submitted, as well as health requirements.
While automation has taken over trivial and labor-intensive positions in other industries, the transition in the healthcare industry has been slow. While appropriate technology exists, there is widespread opposition to replacing ‘hot’ hands with ‘cold’ technology. Things will alter in the future as the share of elderly persons in the population grows. Steps are required to enhance industrial productivity in order to meet the rising demand for health care facilities. Based on prior IT and communications research projects in the health care sector. The utilization of smart home technology and home care visits were both quantified. Smart Home systems are economically viable with current health service organisations, hospitals, and households, but only households are supported. Video tours with rising delivery costs must have an impact on family as well as health services in order to make an inexpensive decision in home care. Because the study is solely quantitative, these findings must be supplemented with qualitative findings and further discussion of the moral, medical, and legal implications of home technology use. Last year, engineering students chose one of the most recent concepts in the computer team for medical devices as their team. Another advantage of the Internet of Things is that the contents of a mobile phone may be accessed from workstations, laptops, or specialised devices. To view these findings, all that is required is a working internet connection. There are several cloud computing services available for viewing these data on the Internet. Sparkfun and IOTGeek are two examples of poorly designed and available services. Tab. 5 compares the suggested video to existing geriatric fall study for various video durations.
Since then, health-care settings have developed research and expertise centered on remote sensor nodes that are technology-oriented. As a result of the underlying cause of heart attacks, as well as a lack of adequate medical care, patients face an unforeseeable risk of death. This is primarily for monitoring the elderly and training doctors and their families. We proposed for a novel approach to reducing accidental mortality rates that employs sensing technology to communicate with loved ones in the event of internet outages caused by the implementation of Patient Wellbeing Tracking. The Deep Belief Networks (DBN) continuous algorithm maps and accumulates information for elderly people using video analysis. This method uses video monitoring to track patients’ well-being. An alarm is transmitted to the patient’s relatives via the appliance if the old patient’s or body’s motions alter abruptly. There are also live patients with Internet timestamps on the screen. As a result, the DC-RBN patient health management system makes extensive use of the Internet to efficiently monitor patient well-being and enable patients to keep track of their lives and jobs. In the future, IoT sensors will be used with deep learning models to increase the closure and detection of senior care in the field of video analysis.
Acknowledgement: We show gratitude to anonymous referees for their useful ideas.
Funding Statement: The authors received no specific funding for this study.
Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the present study.
1. A. Jahanjoo, M. Naderan and M. J. Rashti, “Detection and multi-class classification of falling in elderly people by deep belief network algorithms,” Journal of Ambient Intelligence and Humanized Computing, vol. 11, no. 1, pp. 4145–4165, 2020. [Google Scholar]
2. M. Abdellaoui and A. Douik, “Human action recognition in video sequences using deep belief networks,” Traitement Du Signal, vol. 37, no. 1, pp. 37–44, 2020. [Google Scholar]
3. M. Eldib, F. Deboeverie, W. Philips and H. Aghajan, “Behavior analysis for elderly care using a network of low-resolution visual sensors,” Journal of Electronic Imaging, vol. 25, no. 4, pp. 1–18, 2016. [Google Scholar]
4. M. Uddin, W. Khaksar and J. Torresen, “Ambient sensors for elderly care and independent living: A survey,” Sensors, vol. 18, no. 7, pp. 1–31, 2018. [Google Scholar]
5. D. Lu, Z. He, X. Li, M. Gao, Y. Li et al., “The research of elderly care system based on video image processing technology,” in IEEE 25th Int. Symposium on Industrial Electronics (ISIESanta Clara, CA, USA, pp. 254–258, 2016. [Google Scholar]
6. S. H. Kim and D. W. Kim, “A study on real-time fall detection systems using acceleration sensor and tilt sensor,” Sensor Letters, vol. 10, no. 5–6, pp. 1302–1307, 2012. [Google Scholar]
7. Z. Li-Peng, M. Li-Min, Z. Lei, J. Wei and D. Jian-Ping, “Fall detection algorithm based on bp neural network,” Computer Science, vol. 47, no. 6A, pp. 242–246, 2020. [Google Scholar]
8. G. Vavoulas, M. Pediaditis, E. G. Spanakis and M. Tsiknakis, “The mobifall dataset: an initial evaluation of fall detection algorithms using smartphones,” in IEEE Int. Conf. on Bio Informatics and Bio Engineering, Chania, Greece, pp. 1–4, 2013. [Google Scholar]
9. M. Lu, P. Wei, Z. Yong-Ying, W. Chun-Li and W. Peng-Qian, “Fall action recognition based on deep learning,” Computer Science, vol. 46, no. 9, pp. 106–112, 2019. [Google Scholar]
10. S. V. Kumar, K. Manikandan and N. Kumar, “Novel fall detection algorithm for the elderly people,” in Int. Conf. on Science Engineering and Management Research (ICSEMRChennai, India, pp. 1–3, 2014. [Google Scholar]
11. J. H. He, X. G. Luo, Z. L. Zhang and Y. Yu, “Evolutionary game analysis for the strategic behavior in the provision of elderly care service,” in Chinese Control and Decision Conf. (CCDCHefei, China, pp. 3580–3584, 2020. [Google Scholar]
12. M. Lee, H. Yamazoe and J. H. Lee, “Elderly care training using real-time monitoring system with care training assistant elbow robot (cataro-e),” in Int. Conf. on Ubiquitous Robots (URJeju, Korea (Southpp. 259–264, 2019. [Google Scholar]
13. A. Mercader, M. Biersack, Y. Sun and T. C. Lueth, “A mechanical bed for elderly care to assist while standing, sitting and lying,” in IEEE Int. Conf. on Robotics and Biomimetics (ROBIODali, China, pp. 965–970, 2019. [Google Scholar]
14. K. Yamamoto, S. Shao and N. Kubota, “Heart rate measurement using air pressure sensor for elderly caring system,” in IEEE Symposium Series on Computational Intelligence (SSCIXiamen, China, pp. 1440–1444, 2019. [Google Scholar]
15. M. Qiu, D. Chen, Y. Bian, Z. Sun and Z. Xie, “Research on innovation path of internet plus home-based elderly care service model,” in Int. Conf. on Urban Engineering and Management Science (ICUEMSZhuhai, China, pp. 476–480, 2020. [Google Scholar]
16. Y. Man, L. Ding and Z. Xiaoguo, “Nonlinear system identification method based on improved deep belief network,” in 2018 Chinese Automation Congress (CACpp. 2379–2383, 2018. DOI 10.1109/CAC.2018.8623373. [Google Scholar] [CrossRef]
17. S. Kamada and T. Ichimura, “A video recognition method by using adaptive structural learning of long short term memory based deep belief network,” in IEEE 11th Int. Workshop on Computational Intelligence and Applications (IWCIAHiroshima, Japan, pp. 21–26, 2019. [Google Scholar]
18. B. Qi, Y. Wang, P. Zhang, C. Li and H. Wang, “A novel deep recurrent belief network model for trend prediction of transformer DGA data,” IEEE Access, vol. 7, no. 6, pp. 80069–80078, 2019. [Google Scholar]
19. H. Chen, Y. Zhao, Y. Wang, L. Zhao, L. Yin et al., “A framework of social robot for elderly individuals,” in IEEE Int. Conf. on Consumer Electronics-Taiwan (ICCE-TWYILAN, Taiwan, pp. 1–2, 2019. [Google Scholar]
20. T. F. Bernadus, L. B. Subekti and Y. Bandung, “IoT-based fall detection and heart rate monitoring system for elderly care,” in Int. Conf. on ICT for Smart Society (ICISSBandung, Indonesia, pp. 1–6, 2019. [Google Scholar]
21. B. W. Jeang and Y. Xu, “Intelligent elderly-care in the era of internet and big data,” in IEEE Int. Conf. on Architecture, Construction, Environment and Hydraulics (ICACEHXiamen, China, pp. 59–62, 2019. [Google Scholar]
22. Y. H. Lu and C. C. Lin, “The study of smart elderly care system,” in Eighth Int. Conf. on Information Science and Technology (ICISTCordoba, Granada, and Seville, Spain, pp. 483–486, 2018. [Google Scholar]
23. Y. Taniguchi and H. Matsushita, “A new informatics: Trajectory patterns in health indexes for Japanese elders and systemic change to life course approach,” In: Matsushita H. (eds) Health Informatics. Translational Systems Sciences, vol. 24, Springer, Singapore, 2021. https://doi.org/10.1007/978-981-15-3781-3_4. [Google Scholar]
24. P. S. Mathew and A. S. Pillai, “Boosting traditional healthcare-analytics with deep learning AI: Techniques, frameworks and challenges,” In: Hassanien AE., Taha M.H.N., Khalifa N.E.M. (eds) Enabling AI Applications in Data Science. Studies in Computational Intelligence, vol. 911. Springer, Cham, 2021. https://doi.org/10.1007/978-3-030-52067-0_15. [Google Scholar]
 | This work is licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. |