Research On The Detection Method Of Human's Falling Action

With the development of our country's society and economy,population aging is getting worse,there is an urgent need for automated fall detection.Fall detection methods can be divided into two schools: The first one is wearable device detection method based on physical sensing devices;The second method based on video image processing,which is emerging in recent years.Wearable equipment can accurately obtain the movement parameters of the corresponding joints of the human body,but as a result of its burden of wearing,so it is not very applicable for the elderly.Image processing and pattern recognition have been developing rapidly in recent years,it's time to apply video processing technology to fall detection.Video-based detection method because of its convenience to the audience will have a broad application prospects in the future.At present,many vision-based detection methods follow the method of first location and posterior tracking to obtain human motion.This approach faces four main problems: 1.The computing scale of location and tracking is huge;2.Human body states can only be inferred by obtaining body contour information;3.It's difficlut to loacate the begining of a fall in the video;4.It's easy to follow the wrong target in the tracking process.In order to sovle those four questions,this paper use the video event representation based on sliding window,the outstanding advantage of this approach is to abstract the video content into a spatio-temporal node set.Since the spatio-temporal node is a complete abstraction of the video content,On the one hand,the original location and tracking of the original video can be replaced by the search of the spatio-temporal nodes;On the other hand,content based search can be used instead of inferring human body status through human body contour information.For the spatio-temporal path search problem,we extend the application of the spatio-temporal path search algorithm to fall detection,which was originally appliedto video event detection.By using the algorithm's advantages of the global optimum solution can be obtained in local iterative process and low time complexity,we can solve the problem of how to search for spatio-temporal paths in the condition of unknown starting position fastly.For the problem of path deviation in complex scene detection,the research found that the main reason is that the inter node is lack of predictive guidance during the connection process.According to the characteristics of the fall action.We dismounted the fall action in stages.The Markov process model is used to represent the whole process,and then a transition probability improvement method is proposed to enhance the local connectivity of inter-frame nodes during the spatio-temporal path search process.The transition probability can provid the guidance on the selection of the next frame nodes during the interconnection of inter frame nodes,which solves the problem that the association of inter frame nodes is not strong in the process of path search.In order to verify the validity of the algorithm,we produced a new multi-scene fall data set,and validated the above algorithm on the data set.Experimental results show that the proposed method can be effectively applied to the data set and can achieve better detection results.