Dissertations / Theses on the topic 'Human Fall detection'

L’ Ambient Intelligence rappresenta la sfida del futuro. Per ottenere un ecosis- tema funzionante e calibrato in base alle esigenze dell’utente, necessario inte- grare numerosi sistemi, ciascuno dedicato a un compito specifico. Uno di questi sottosistemi il rilevamento della caduta umana. In questa tesi, il rilevamento delle cadute umane affrontato da una prospettiva audio. In questo lavoro stato presentato un set di dati denominato A3FALL, composto da una serie di diversi eventi audio relativi alla caduta di oggetti comuni e cadute umane, simulate e reali. In particolare, per tale scopo stato sviluppato uno speciale sensore acus- tico a pavimento (FAS) che stato utilizzato per registrare il set di dati insieme ad un array di microfoni. Sono stati proposti differenti approcci che lavorano con una diversa base di conoscenze in base al compito specifico: in primo lu- ogo sono stati descritti due approcci supervisionati che hanno evidenziato le peculiarit del rilevamento della caduta audio e hanno dimostrato l’efficacia del sensore proposto. Per via delle poche cadute umane reali a disposizione per lo sviluppo di tali sistemi, sono stati proposti approcci non supervisionati che non necessitano di esempi della classe target nella fase di apprendimento. stato di- mostrato che gli approcci non supervisionati hanno prestazioni migliori rispetto ai sistemi dello stato dell’arte, ma funzionano bene in scenari poco complessi. Infine, sono stati sviluppati e descritti metodi che funzionano in condizioni pi realistiche. Viene proposto un sistema in cui l’utente interviene correggendo il funzionamento del sistema per una notevole riduzione dei falsi allarmi. Poi viene proposto un approccio di apprendimento one-shot che, senza l’intervento dell’utente, pu ottenere risultati promettenti utilizzando solo alcuni esempi di caduta umana nella fase di addestramento. La tesi si conclude con una valu- tazione approfondita di un approccio basato su un Autoencoder siamese. stato dimostrato che questo approccio migliore di tutti i sistemi precedentemente proposti quando valutatti in scenari complessi.
At present, Ambient Intelligence represents the challenge of the future. To obtain an ecosystem that is fully functional and calibrated to the user need, numerous systems, each of them dedicated to a specific task, must be integrated. One of these sub-systems is the human fall detection. Both research community and governments gave particular attention to the human fall detection because the fall is the first cause of death for people over 65. In this thesis, the human fall detection is addressed from an audio perspective: a dataset named A3FALL, composed of a corpus of several audio fall events of every-day objects and both simulated and real human falls recorded in 3 different rooms, has been presented. In particular, a special floor acoustic sensor (FAS) has been developed from this purpose and used to record the dataset together with an array of a microphone array. Different approaches that work with a different knowledge base according to the specific task have been proposed: first, two supervised approaches have been described that have highlighted the peculiarities of the audio drop detection and demonstrated the effectiveness of the proposed sensor. The human falls hardly available for systems development, unsupervised systems have been proposed that do not need examples of the target class in the learning phase. It has been shown that unsupervised approaches have better performance than the art state systems, but they do work well in not very complex scenarios. Finally, methods that work under more realistic conditions have been developed and described. A system where the user intervenes by correcting the system’s operation for a considerable reduction of false alarms is proposed. Then a few-shot learning approach that without any user intervention can achieve promising results using only a few examples of human fall in the training phase has been presented. The thesis concludes with an extensive evaluation of a Siamese Convolutional Autoencoder based approach. It has been shown that this approach outperforms all the previously proposed systems when assessed in a complex scenario.

Wearable sensor systems composed of small and light sensing nodes have the potential to revolutionise healthcare. While uptake has increased over time in a variety of application areas, it has been slowed by problems such as lack of infrastructure and the functional capabilities of the systems themselves. An important application of wearable sensors is the detection of falls, particularly for elderly or otherwise vulnerable people. However, existing solutions do not provide the detection accuracy required for the technology to gain the trust of medical professionals. This thesis aims to improve the state of the art in automated human fall detection algorithms through the use of a machine learning based algorithm combined with novel data annotation and feature extraction methods. Most wearable fall detection algorithms are based on thresholds set by observational analysis for various fall types. However, such algorithms do not generalise well for unseen datasets. This has thus led to many fall detection systems with claims of high performance but with high rates of False Positive and False Negative when evaluated on unseen datasets. A more appropriate approach, as proposed in this thesis, is a machine learning based algorithm for fall detection. The work in this thesis uses a C4.5 Decision Tree algorithm and computes input features based on three fall stages: pre-impact, impact and post-impact. By computing features based on these three fall stages, the fall detection algorithm can learn patterns unique to falls. In total, thirteen features were selected across the three fall stages out of an original set of twenty-eight features. Further to the identification of fall stages and selection of appropriate features, an annotation technique named micro-annotation is proposed that resolves annotation-related ambiguities in the evaluation of fall detection algorithms. Further analysis on factors that can impact the performance of a machine learning based algorithm were investigated. The analysis defines a design space which serves as a guideline for a machine learning based fall detection algorithm. The factors investigated include sampling frequency, the number of subjects used for training, and sensor location. The optimal values were found to be10Hz, 10 training subjects, and a single sensor mounted on the chest. Protocols for falls and Activities of Daily Living (ADL) were designed such that the developed algorithms are able to cope under a variety of real world activities and events. A total of 50 subjects were recruited to participate in the data gathering exercise. Four common types of falls in the sagittal and coronal planes were simulated by the volunteers; and falls in the sagittal plane were additionally induced by applying a lateral force to blindfolded volunteers. The algorithm was evaluated based on leave one subject out cross validation in order to determine its ability to generalise to unseen subjects. The current state of the art in the literature shows fall detectors with an F-measure below 90%. The commercial Tynetec fall detector provided an F-measure of only 50% when evaluated here. Overall, the fall detection algorithm using the proposed micro-annotation technique and fall stage features provides an F-measure of 93% at 10Hz, exceeding the performance provided by the current state of the art.

Human Fall Detection is a research area with interest from many disciplines and aims to perform for many assisted-living monitoring applications to promptly identify life-threatening situations. A fall occurs when a person is unable to maintain balance due to a variety of issues; physical; mental or environmental. The accurate detection of the fall is crucial as a missed detection can be fatal. Variability of human physiological characteristics is currently unstudied as to the impact on a fall detector's performance as young adults and elderly are expected to fall differently. Another important issue is the scene occlusions. In the use of visual sensors, an occluded fall is treated as a missed detection as the whereabouts of the person is unknown when occluded. Finally, current studies are based on acted fall datasets on which algorithms are trained. These dataset are unrepresentative of real fall events and illustrate the events without occlusions or other scene in uences. Several fall detection algorithms were developed during the study aiming to achieve accuracy in detection falls while fall-like actions such as lying down remain undetected. Human fall datasets were used for training and testing purposes of A machine learning algorithm using data from depth cameras which captured the fall events from different views. A new pathway was introduced tackling the issues of availability issues of data-driven machine learning approaches which was achieved with the use of simulation data. The use of myoskeletal simulation was then selected as a closer representation of the human body in terms of structure and behaviour. With the use of a simulation model, a personalised estimation of the fall event can be achieved as it is parametrised on a physical characteristic such as the height of the falling person. Alternative technologies such as accelerometers have been used for fall detection to prove the validity of this approach on other modalities. A study regarding the impact of occlusions for fall detection which is one of the issues not properly investigated in current work is proposed and examined. Synthetic occlusions were added to existing depth data from publicly available datasets. The research methodologies were evaluated using the most representative depth video and accelerometer data from existing datasets, as well as YouTube videos of real-fall events. The machine learning methodologies achieved good results on similar body variability datasets. A discussion regarding the proof of concept of the simulation-based approach for fall modelling is mentioned given the comparative results against existing methodologies which achieves better than any existing work evaluated against known datasets. The simulation approach is also evaluated against occluded fall and non-fall event data, proving the further robustness of the approach. This platform can be expanded to analyse any type of fall, or body posture (e.g. elderly), without the use of humans to performs fall events.

The research in this thesis is intended to aid caregivers’ supervision of toddlers to prevent accidental injuries, especially injuries due to falls in the home environment. There have been very few attempts to develop an automatic system to tackle young children’s accidents despite the fact that they are particularly vulnerable to home accidents and a caregiver cannot give continuous supervision. Vision-based analysis methods have been developed to recognise toddlers’ fall risk factors related to changes in their behaviour or environment. First of all, suggestions to prevent fall events of young children at home were collected from well-known organisations for child safety. A large number of fall records of toddlers who had sought treatment at a hospital were analysed to identify a toddler’s fall risk factors. The factors include clutter being a tripping or slipping hazard on the floor and a toddler moving around or climbing furniture or room structures. The major technical problem in detecting the risk factors is to classify foreground objects into human and non-human, and novel approaches have been proposed for the classification. Unlike most existing studies, which focus on human appearance such as skin colour for human detection, the approaches addressed in this thesis use cues related to dynamic motions. The first cue is based on the fact that there is relative motion between human body parts while typical indoor clutter does not have such parts with diverse motions. In addition, other motion cues are employed to differentiate a human from a pet since a pet also moves its parts diversely. They are angle changes of ellipse fitted to each object and history of its actual heights to capture the various posture changes and different body size of pets. The methods work well as long as foreground regions are correctly segmented.

The deep learning technique Human Pose Estimation (or Human Keypoint Detection) is a promising field in tracking a person and identifying its posture. As posture and balance are two closely related concepts, the use of human pose estimation could be applied to fall prevention. By deriving the location of a persons Center of Mass and thereafter its Center of Pressure, one can evaluate the balance of a person without the use of force plates or sensors and solely using cameras. In this study, a human pose estimation model together with a predefined human weight distribution model were used to extract the location of a persons Center of Pressure in real time. The proposed method utilized two different methods of acquiring depth information from the frames - stereoscopy through two RGB-cameras and with the use of one RGB-depth camera. The estimated location of the Center of Pressure were compared to the location of the same parameter extracted while using the force plate Wii Balance Board. As the proposed method were to operate in real-time and without the use of computational processor enhancement, the choice of human pose estimation model were aimed to maximize software input/output speed. Thus, three models were used - one smaller and faster model called Lightweight Pose Network, one larger and accurate model called High-Resolution Network and one model placing itself somewhere in between the two other models, namely Pose Residual Network. The proposed method showed promising results for a real-time method of acquiring balance parameters. Although the largest source of error were the acquisition of depth information from the cameras. The results also showed that using a smaller and faster human pose estimation model proved to be sufficient in relation to the larger more accurate models in real-time usage and without the use of computational processor enhancement.
Djupinlärningstekniken Kroppshållningsestimation är ett lovande medel gällande att följa en person och identifiera dess kroppshållning. Eftersom kroppshållning och balans är två närliggande koncept, kan användning av kroppshållningsestimation appliceras till fallprevention. Genom att härleda läget för en persons tyngdpunkt och därefter läget för dess tryckcentrum, kan utvärdering en persons balans genomföras utan att använda kraftplattor eller sensorer och att enbart använda kameror. I denna studie har en kroppshållningsestimationmodell tillsammans med en fördefinierad kroppsviktfördelning använts för att extrahera läget för en persons tryckcentrum i realtid. Den föreslagna metoden använder två olika metoder för att utvinna djupseende av bilderna från kameror - stereoskopi genom användning av två RGB-kameror eller genom användning av en RGB-djupseende kamera. Det estimerade läget av tryckcentrat jämfördes med läget av samma parameter utvunnet genom användning av tryckplattan Wii Balance Board. Eftersom den föreslagna metoden var ämnad att fungera i realtid och utan hjälp av en GPU, blev valet av kroppshållningsestimationsmodellen inriktat på att maximera mjukvaruhastighet. Därför användes tre olika modeller - en mindre och snabbare modell vid namn Lightweight Pose Network, en större och mer träffsäker modell vid namn High-Resolution Network och en model som placerar sig någonstans mitt emellan de två andra modellerna gällande snabbhet och träffsäkerhet vid namn Pose Resolution Network. Den föreslagna metoden visade lovande resultat för utvinning av balansparametrar i realtid, fastän den största felfaktorn visade sig vara djupseendetekniken. Resultaten visade att användning av en mindre och snabbare kroppshållningsestimationsmodellen påvisar att hålla måttet i jämförelse med större och mer träffsäkra modeller vid användning i realtid och utan användning av externa dataprocessorer.

Fallolyckor bland äldre är ett stort folkhälsoproblem. Således har det utvecklats en rad olika system för fjärrövervakning av äldre för att möjliggöra en tidig detekte- ring av fallolyckor. Majoriteten av de forskningar som hittills har gjorts inom fallo- lyckor har fokuserat på att utveckla nya mer framgångsrika algoritmer just för att identifiera fall från icke-fall. Trots det visar statistiken att dödsfall och skador orsa- kade av fallolyckor ökar för varje år i samband med den ökande andel äldre perso- ner i befolkningen. Detta examensarbete handlar om att förbättra nuvarande falldetekteringsanord- ningar genom att täcka de brister och tillgodose de behov som finns i nuvarande tekniker. De förbättringar som har kartlagts är att kunna ge en säkrare bedömning av patientens hälsa och kunna påkalla snabbare hjälpinsats när en fallolycka har inträffat. En annan förbättring är rörelsefriheten för äldre. De ska kunna befinna sig utomhus och ha möjlighet att utföra sina dagliga aktiviteter utan att vara be- gränsade av platspositionen. Sammanfattningsvis kan det konstateras att ett multisensorsystem i form av en prototyp har konstruerats för att täcka de brister som har kartlagts. Utöver att pro- totypen detekterar fall och kroppsrörelser med hjälp av en accelerometersensor innehåller den även en sensor för detektering av livstecken i from av EKG. Den in- nehåller dessutom cellulära och trådlösa nätverksanslutningar i form av GPRS och Wi-Fi för att möjliggöra rörelsefriheten hos äldre. Vidare innehåller prototypen en sensor för GPS som ger information om platsposition.
Accidental falls among the elderly is a major public health problem. As a result, a variety of systems have been developed for remote monitoring of the elderly to permit early detection of falls. The majority of the research that has been done so far in fall accidents has focused on developing new more successful algorithms spe- cifically to identify fall from non-fall. Although the statistics show that mortality and injuries caused by falls are increasing every year in conjunction with the in- creasing proportion of older people in the population. This thesis is about improving the current fall detection devices by covering the gaps and meet the needs of the current fall detection techniques. The improve- ments that have been identified is to provide a secure assessment of the patient's health and be able to call for aid more quickly when a fall occurs. Another im- provement is the mobility for the elderly to be outdoors and have the ability to per- form daily activities without being limited by the location position. In summary it can be said that a multisensor system in form of a prototype has been designed to cover the deficiencies and improvements that have been identi- fied. Apart from detection of falls and body movements through an accelerometer sensor the prototype does also include a sensor for detecting vital signs in form of ECG. It also supports cellular and wireless network communication in form of GPRS and Wi-Fi to enable freedom of movement for the elderly. Furthermore, the prototype includes a sensor for GPS that provides information about location position.

Le vieillissement de la population est l'un des problèmes clés pour la grande majorité de nombreux pays. Le nombre de personnes âgées souffrant de multiples maladies et nécessitant une surveillance continue de leurs signes vitaux augmente chaque jour, entraînant des coûts de santé supplémentaires. Les systèmes de santé modernes en médecine gériatrique nécessitent souvent la présence de personnes âgées à l'hôpital, ce qui est en conflit avec leur exigence d'indépendance et d'intimité. Les développements récents sur la télésurveillance e-santé offrent une large gamme de solutions. Cependant, la plupart des appareils sont conçus pour une détection médicale spécifique et fonctionnent indépendamment les uns des autres. Il y a toujours un manque de cadre intégré avec une interopérabilité élevée et un support de surveillance en ligne continu pour une analyse de corrélation plus approfondie. Cette thèse est une étape vers un système de collecte de données à distance, complet et continu pour les personnes âgées présentant divers types de problèmes de santé. Notre esprit de recherche est motivé par la demande immédiate d'un système de surveillance de la santé à distance sans fil sécurisé et fiable pour les personnes âgées en résidence assistée, combinant diverses sources de données. Pour créer un système aussi complet, nous le divisons en sous-systèmes, afin de le rendre réalisable et facile à mettre en œuvre, nous permettant ainsi de mettre à jour chaque sous-système individuellement dans les études futures sans affecter les autres sous-systèmes intégrés. L'accent est mis sur un système complet de surveillance à distance de l'e- santé. La liste des principales contributions contient (1) proposer une nouvelle approche pour la sécurité des appareils surveillés et proposer une solution pour prévenir les attaques MiTM et réduire la consommation d'énergie, (2) nous proposons une détection de chute fiable, (3) étudier et développer une nouvelle méthode de reconnaissance des activités quotidiennes des patients âgés surveillés, (4) proposer une approche pour améliorer la fiabilité du système et réduire les fausses alarmes et les interventions inutiles, (5) proposer et développer un algorithme de conversion de la langue des signes en texte utilisant une analyse de fusion multi-capteurs. En conséquence, nous prévoyons de fournir un système de surveillance avec une précision fiable dans la détection d'événements anormaux et de déclencher une alarme lors de la détection de tels événements pour demander de l'aide et de l'assistance
Aging population is one of the key problems for the vast majority of many countries. The number of elderly people who suffer from multiple diseases and need continuous monitoring of their vital signs increases everyday, resulting in additional healthcare costs. Modern healthcare systems in geriatric medicine often require elderly presence at the hospital which conflict with their demand for independence and privacy. Recent developments on remote e-health monitoring, provides a wide range of solutions. However, most of the devices are designed for specific medical sensing and operate independently from each other. There is still a lack of integrated framework with high interoperability and continuous online monitoring support for further correlation analysis. This thesis is a step towards a remote, complete, and continuous data gathering system for elderly people with various types of health problems. Our research spirit is motivated by immediate demand in a secured and trusted remote wireless health monitoring System for assisted living Elderly people, combining various data sources. To create such a complete system we divide it into subsystems, in order to make it feasible and easy to implement, thus allowing us to update each subsystem individually in the future studies without affecting other integrated subsystems. The main focus is on a complete remote e-health monitoring system. The list of main contributions contains (1) propose a new approach for security of monitored devices and propose a solution to prevent MiTM attacks and reduce energy consumption, (2) we propose reliable fall detection,(3) investigating and developing a novel recognition method of daily activities for monitored elderly patient, (4) propose an approach to enhance the reliability of the system and to reduce false alarms and unnecessary interventions, (5) propose and develop a sign language to text converter algorithm using multi-sensor fusion analysis. As a result, we expect to provide a monitoring system with reliable accuracy in the detection of abnormal events, and raise an alarm upon detection of such events to seek help and assistance

碩士
國立中正大學
電機工程所
96
In recent years, because the decline in birth rate and the increase of national average life span, special attention is gradually shifting to the security of the elderly people and children, fall is one of the often seen accidental among them. The purpose of this thesis is to detect the situation and to examine the possibility whether or not the personages fall while moving by a monitoring system. Based on a result of detection, the doctor and medical treatment-nursing group could put forward a medical treatment plan and carry on improvement of the pedestrian''s activity environment. The application of computer vision systems has widely spread along with the progress of the computer software and hardware. In the recent ten years, the researchers show their interest in the research of identifying and analyzing the humans'' motions or behaviors. The academia has a quite good result of study in the distinguishing of human motion posture at present. This thesis proposed a detection system on examining and judging of the pedestrian who falls for the abnormal behavior or personage''s own independence. This thesis mainly divides into two focal points: the fist is individual distinguish personages and the second is the detection of fall. Firstly, the personage''s information and definition of the upper part of the body information were found, this result was carry on and be the method for human tracking. Due to the complexional information of human, the group of other moving objects could be leach out. Besides, the part of fall detection, we detect the person in video who falls based on the information of its horizontal, vertical projections. The function of this system can be used on human tracking and fall detection, and it could be achieved without define models and large number of mathematics operation.

碩士
國立臺灣科技大學
電子工程系
102
In recent years, the global population has begun to age rapidly. Automatic fall detection for senior citizens has become an important issue for smart home. In this research, we propose a novel video-based human fall detection system that can detect a human fall in real-time with a high detection rate. This fall detection system is based on Hidden Conditional Random Fields model, and an intelligent combination of height estimation and appearance cues. Our system can efficiently distinguish “fall-down incidents” from “fall-like incidents” such as sit-down and squat. Experimental results indicate that the proposed human fall detection system can achieve a high detection rate and low false alarm rate. Also, the proposed system outperforms Hidden Markov Chain and Cuboids in terms of detection rate.

碩士
國立中正大學
電機工程研究所
106
In recent years, due to the development of medical technology, people are more and more long-lived, and proportion of the elderly population has increased year by year. Therefore, the medical care of the elderly is becoming more and more important. We used computer vision and image processing technology to surveil the medical care of the elderly family. In this paper, we use two cameras to capture different views in the same scene for abnormal behavior events (such as falling down) detection and identification. This paper combines long-term 2D human contour to establish and identify abnormal behavior models with deep learning architecture. The anomalous behavior recognition algorithm proposed in this paper consists of three main steps. First, we detect 2D human contour from surveillance cameras. Second, we use motion history image (MHI) method and combine long-term sequence actions into several MHI images. Finally, we use deep learning technology (CNN and CNN + LSTM architectures) to recognize the abnormal behavior of MHI sequences. This method not only recognizes the motions of walking, standing, falling down, but recognizes rising to avoid excessive false alarms. This paper compared different segments of the abnormal behavior identification. Due to the falling frames are about 3-5 frames, the abnormal behavior (falling down) recognition rate will be low if we didn't divide the sequences in long-time recognition. We compared the performance of abnormal behaviors that used different training methods and different ground truth (GT) of the same deep learning architecture (CNN + LSTM architectures). We also compared the performance that used different architectures (CNN architectures and CNN + LSTM architectures) and single/double cameras. In this paper, we experienced the method that used two-stage training and GT was the majority decision. We got the accuracy of abnormal behaviors identification is 97.66%.

碩士
國立臺灣科技大學
自動化及控制研究所
107
In recent years, due to medical progress, human life has become longer and longer, coupled with the phenomenon of “less child” in the country, causing the population to age. The home-cared issue is often on the news page, and the fall is easy to cause damage to the elderly. Therefore, we use computer vision. And deep learning techniques to detect the inconvenience of movement or the activity status of the elderly, and to solve the problem of insufficient manpower in the nursing center. In view of the fact that previous theses often use manual tags to define the parameters of falls. This thesis uses two-dimension image vision technology with deep learning algorithms to propose a system that can automatically identify and detect falls of the elderly. The system structure of this thesis can be divided into three parts. Firstly, the image information of the moving object is obtained by Dense Optical Flow. Then use the Sliding Window technology to divide the big movie into many small video clips, each of clips have 10 images. Finally, the LRCN (CNN+RNN) hybrid model in depth learning technology is used to analyze the optical flow image and discriminate the occurrence of the fall event. The CNN extraction feature model uses the VGG16 architecture, while the RNN timing discrimination model uses the LSTM architecture as the hidden layer. Finally, we tune the hyperparameters and then use five cross-validation to check whether the model has overfitting problems or not. The experimental results show that the identification of the fall event can achieve sensitivity 99.76693% and accuracy 97.67701%.

碩士
國立臺灣科技大學
電子工程系
100
We present a novel real-time video-based human fall detection system in this thesis. Because the system is based on a combination of shadow-based features and various human postures, it can distinguish between fall-down and fall-like incidents with a high degree of accuracy. To support effective operation in different viewpoints, we propose a new feature called virtual height that can estimate the body height without 3D model reconstruction. As a result, the model is low computational complexity. Our experiment results demonstrate that the proposed system can achieve a high detection rate and a low false alarm rate.

碩士
中原大學
電子工程研究所
96
In recent years, the number of the elderly increases rapidly. Consequently the health care of the elderly attracts more and more attention and we need to invest a lot of resources to maintain its quality. This thesis proposes a vision-based fall detection system using an omni-directional camera for the elderly and patients at home and in health-care institutions, hoping to detect the fall accident immediately when it happens and notify the medical personnel to provide the emergency care in time. In order to make the system more practical in real environments, we consider the practical environmental factors that may take place in our daily life. For instance, the occurrence of light source glimmer and turning a light on and off and leaving over static abandoned objects in the environment and resulting in multiple targets. The former can be solved by detecting the degree of luminance changes and the latter can be solved by using the static characteristic of abandoned objects. In addition we divide the fall down patterns in omni-directional images into non-radial and radial directions according to the angle associated with a body line. We further categorize the radial fall down patterns into inward and outward directions. We extract suitable features for these three different fall down patterns, including angle and length variation associated with the body line and Motion History Images. Given these features, a simple thresholding and decision tree technique is adopted for fall detection. Experimental results show that the proposed system has overcome the practical environmental factors of light source glimmer and static abandoned objects, increasing the practicability of the system in a real-world environment. In fall-down detection, since multiple fall-down patterns are considered, the recognition accuracy of the fall down system is improved from 0.73 to 0.87 and the Kappa value is improved from 0.47 to 0.75. These results show that we have proposed an effective fall down detecting system in this thesis.

碩士
中華大學
電機工程學系
106
In recent years, the aging society has become a common phenomenon in all countries. There is no exception in Taiwan. Therefore, the care of the elderly is becoming more and more important. In the elderly accident, falling is the second most important factor. The occurrence of falls often causes fractures, and causes physical and mental injury. It can be seen that falls are a threat to the health of the elderly. With the performance of the computer is getting faster, CCD costs are getting lower and lower, enabling people to analysis of real-time images through image processing. This paper proposes a fall detection system that uses OpenPose[12] to extract skeletons through a camera. It is mainly used to detect human fall. This system can detect and judge whether a pedestrian falls in a multi-person environment.

碩士
國立清華大學
電機工程學系
97
Fall and slip of the elderly are the main concerns for home care or day care center. We proposed a method to detect a slip event and a fall event by computing integrated spatiotemporal energy (ISTE) map that includes motion and time of motion occurrence as our motion feature. The extracted human shape is represented by an ellipse that provides crucial information of human motion activities. We use this features to detect the events in the video with non-fixed frame rate. In this work, we assume that the person is on the ground with no or little motion after the fall accident. Our experiments are demonstrated in the indoor and the outdoor scenes, where the illuminate condition varies. So the threshold of the background subtraction and the parameters of the smoothing filter are adjusted independently. Experimental results show that our method is effective for fall and slip detection. The total number of testing frames is about , and we use an Intel Core2 Duo 1.8GHz CPU on a Microsoft Windows XP operating system.

碩士
中原大學
電子工程研究所
94
Due to the advancement of technology and medicine, people begin to pay more attention to the quality improvement of health care. Many researches show that the fall accident occupies 80% of all accidents in a hospital. The fall accident may cause the condition of a patient deteriorated, producing complications and extending the patient’s stay in the hospital. As a result, it increases the burden of a family and seriously wastes medical resources from the society. Thus, preventing the fall accident and detect it immediately is one of the important topics regarding the quality improvement of health care. This thesis proposes a reliable tele-care system that can detect the fall accident immediately, notify medical personnel when the accident occurs, prevent the patient’s condition from deteriorating due to late treatment, and reduce the burden of medical personnel. 　A unique feature of the proposed system is that we use a MapCam to capture 360∘scense simultaneously and eliminate any blind spot. Furthermore, personal information is integrated into the system and makes it smarter by customizing the system for each individual. With personal information (including basic personal data, danger factor, electronic health history, etc), we can adjust the detection sensitivity on a case by case basis to reduce unnecessary alarms, and put more attention on the elderly with special diseases or conditions. We also propose another fall detection algorithm for various falling directions and walking paths. The experimental results show that using a simple fall detection algorithm and combining it with simple personal information can raise fall detection accuracy and reliability effectively in a particular environment. When the algorithm itself is robust enough, perhaps the detection accuracy can be increased only if biomedical signals are considered as well. The experimental results also show that the new fall detection algorithm proposed here can do a good job in an indoor environment for all fall cases (different walking paths and falling directions). 　The successful recognition rate and kappa value of our system with personal information are 0.92 and 0.92, respectively, showing that we have a reliable system.

碩士
國立交通大學
多媒體工程研究所
96
In the province of Human Computer Interaction, monitor system is an important study. As long as the situation of aging society becomes more and more serious, the care costs will increase plenty. That is the reason so many domestic and foreign scholars throw themselves into the research of elderly care monitor system in order to support the existing care system and reduce the huge expenditures of labor costs. This research used and integrated the human face detection system developed by our laboratory to get the characteristics of the human body and track that. And also used k-th Nearest Neighbor classification to classify the human postures. Then using the information of the changing rate collected by many experiments this research finally can develop a fall detection system.

碩士
國立交通大學
生醫工程研究所
106
The issue of health care of elder is more important in recent years. Therefore, the research of fall detection system is very flourishing. In this thesis, we propose a fall detection system based on image processing. This system can detect falls of people in surveillance environment by motion and human shape in image. When system detect fall, it will send message to reduce waiting time and increase the survival rate. There are two contributions of this thesis. First, we proposes a set of features by foreground and human shape. Train a classifier make the result of fall detection system more accurately. Finally, we completes the prototype of system in fall detection base on our proposed feature and motion history image, and the system can distinguish between falls and normal activities such as sitting, lying or crouching. From experimental results, our proposed method can increase efficacy of fall detection. We also prove our system can detect fall in several continuous movement by video. Finally, we integrate our proposed system with accelerometer to increase the accuracy of fall detection system.

Yes
Human fall detection systems play an important role in our daily life, because falls are the main obstacle for elderly people to live independently and it is also a major health concern due to aging population. Different approaches are used to develop human fall detection systems for elderly and people with special needs. The three basic approaches include some sort of wearable devices, ambient based devices or non-invasive vision-based devices using live cameras. Most of such systems are either based on wearable or ambient sensor which is very often rejected by users due to the high false alarm and difficulties in carrying them during their daily life activities. This paper proposes a fall detection system based on the height, velocity and position of the subject using depth information from Microsoft Kinect sensor. Classification of human fall from other activities of daily life is accomplished using height and velocity of the subject extracted from the depth information. Finally position of the subject is identified for fall confirmation. From the experimental results, the proposed system was able to achieve an average accuracy of 94.81% with sensitivity of 100% and specificity of 93.33%.
Partly sponsored by Center for Graduate Studies. This work is funded under the project titled “Biomechanics computational modeling using depth maps for improvement on gait analysis”. Universiti Tun Hussein Onn Malaysia for provided lab components and GPPS (Project Vot No. U462) sponsor.

A área da saúde foi uma das muitas beneficiadas com a evolução tecnológica, dando origem a novos conceitos que visam melhorar ou mesmo prolongar a vida das pessoas. Os sistemas de monitorização vestíveis, juntamente com as comunicações sem fios, são a base de uma classe emergente de redes de sensores. Estas tecnologias de informação permitem a deteção precoce de condições anormais e ajudam na sua prevenção. O objetivo é criar um destes sistemas compostos por uma rede de sensores que é implementada numa peça de roupa através de fios condutores com sensores conectados. Em contato com o corpo humano tem a função de fazer várias leituras, e.g., temperatura corporal, pulsação, entre outras. Outro objetivo é detetar quedas do utilizador. A deteção de quedas é cada vez mais importante para o utilizador, pois é uma situação que pode colocar em risco a sua saúde. Para o desenvolvimento deste conceito, são utilizadas Comunicações Móveis e o Sistema de Posicionamento Global. A primeira é uma tecnologia que permite criar chamadas de emergência em resposta a alarmes do sistema, o segundo indica qual a sua posição geográfica. Para complementar o sistema, existe uma plataforma online que regista a posição do utilizador tal como os seus dados. Tem também uma área de alertas no qual o utilizador pode verificar os seus valores preocupantes. Em caso de emergência o sistema contacta os serviços de emergência ou em casos especiais a ajuda pode ser obtida através de um UAV.
The health area was one of the many beneficiaries of technological evolution, giving rise to new concepts that aim to improve or even prolong people’s lives. Wearable monitoring systems, along with wireless communications, form the basis of an emerging class of sensor networks. These information technologies enable the early detection of abnormal conditions and help in their prevention. The goal is to create one of these systems composed by a network of sensors that is implemented in a garment through conductive wires with connected sensors. In contact with the human body it has the function of doing several readings, e.g., body temperature, heartbeat, among others. Another goal is to detect user falls. The detection of falls is increasingly important for the user, as it is a situation that can endanger people’s health. For the development of this concept, Mobile Communications and the Global Positioning System are used. The first is a technology that allows to create emergency calls in response to system alarms, the second indicates the geographical location. To complement the system there is an online platform that registers the position of the user as well as his data. There is also an alert area in which the user can check his alarming values. In case of emergency the system contacts the emergency services or in special cases help can be obtained through an UAV.

碩士
長庚大學
醫療機電工程研究所
98
Preventing Falling down is an important issue for the aging society; therefore, the fall detection is crucial for the healthcare system. Fall signal is generally collected from a 3-axis accelerometer which is placed on the human’s chest, and the collected signal is further analyzed to develop the fall detection algorithms. Nevertheless, it is not easy to collect realistic fall signals because the experiments of falls may cause serious injuries. Therefore, most of collected fall signals are conservative and cannot completely represent the situations of actual falls. That means the volunteer may perform a slow fall when collecting the fall signal. Especially, it is hardly to collect the fall signals from high risk fall situations such as falls from stairs. Biped humanoid robot researches are fast increasing in recent years, because the torso structures of the biped humanoid robots is similar to the human beings. This thesis proposes a biped humanoid robot based fall scenario simulation system. The proposed fall scenario simulation system constructs the gait pattern libraries for the different fall scenarios which are similar to the falls of human beings. A 3-axis accelerometer is also placed on the chest of the biped humanoid robot to measure the fall signals. In order to verify the proposed approach, a motion capture system is employed in this study to measure the fall motions. At the same time, the fall motions collected from the motion capture system and the fall signals collected from the 3-axis accelerometer are synchronously recorded to verify the signal correlations between the biped humanoid robot and the human beings. Experiment results shows that the signal correlations between the biped humanoid robot and the human beings for typical forward, side and backward falls. Based on this correlation performance, the high risk fall signals such as falls from stairs and slip falls are collected from the biped humanoid robots only. Therefore, the proposed biped humanoid robot based fall scenario simulation system may effectively collect the fall signals for the further fall detection algorithm studies.