Research On Fall Detection Algorithm And System Implementation On Mobile Robot Platform

With the aggravation of population aging,countries around the world are facing social problems of increasing elderly population and tight medical resources.Among them,falls monitoring and ambulance for the elderly is one of the key points.Research institutes and enterprises all over the world have carried out extensive research and experiments on fall detection schemes.Many fall detection algorithms and commercial products have come out one after another,but there are generally problems such as low accuracy,limited practicality,and poor user experience.With the development of artificial intelligence and automation teclhnology,mobile robots play an increasingly important role in various fields,as well as in fall detection tasks.While improving user experience and enhancing human-computer interaction,mobile robots can make up for the lack of flexibility of traditional detection schemes.Based on this,this thesis aims to study and implement a fall detection algorithm based on mobile robots,improve the accuracy of the algorithm and improve the user experience in real life scenarios.The main work of this thesis are summarized as follows:(1)The advantages and disadvantages of various fall detection algorithms at home and abroad are discussed,and the daily behavior of the elderly is summarized.The changes of human posture characteristics during and after a fall are analyzed in detail.The fall movement is defined.It is pointed out that the detection algorithm can be designed and implemented according to the motion characteristics of human body after a fall and the semantic relationship between human and the ground.(2)A fall detection algorithm based on human contour features is designed and implemented,which takes advantage of the large variation of human contour features during the fall process.Firstly,the obj ect detection algorithm is used to detect the body area,and then extrace the four features of the area,such as width-height ratio,centroid height,off-ground height and direction angle.The features are used to train the SVM classifier for classification.The accuracy of the algorithm on public data sets is higher than that of the existing algorithms.(3)A new detection algorithm CRN(Coexistence Relation Network)is proposed.CRN not only extracts the details of the object itself,but also effectively utilizes the global context information to assist the detection task,so as to improve the accuracy of the detection algorithm.In addition,CRN algorithm is end-to-end,which can simplify the training and testing process and ensure that the algorithm can complete the detection task quickly and accurately.The results show that CRN structure can improve the mAP of object detection by about 1%,action recognition by 1.3%and fall detection by 1.7%.(4)A fall detection system based on mobile robots is built to verify the effectiveness and practicability of the proposed algorithm under limited computing resources.Mobile robots are responsible for image acquisition,transmission and location adjustment.Cloud detection platform is introduced to complete the tasks of receiving data,falling detection and alarm.In real life scenarios,the whole system can smoothly complete the fall detection task and has a better user experience.