Research On Indoor Tracking And Behavior Detection Of The Elderly Living Alone Based On KCF

With the aging problem becoming more and more serious in China and even in the world,the safety of the elderly living alone at home has become an urgent problem to be solved in today's society.The traditional solution is to use tedious wearing equipment to detect the daily behavior of the elderly.In order to solve the shortcomings of traditional methods,the use of intelligent surveillance video has important significance for the behavior detection of indoor elderly living alone.In order to detect and analyze the behavior of the elderly living alone indoors,the first thing we need to do is to detect and track the indoor target,then based on the target feature extraction,we need to analyze and judge the target and perform behavior recognition and classification.The main work is as follows:(1)Moving target detection.ViBe algorithm is simple and fast to detect moving objects,but because it uses the change of pixels to detect objects,the extracted moving objects are not complete enough.In order to solve the problem of ghost or residual shadow and target void in ViBe algorithm,the method of choosing neighborhood is changed when building model.When using morphological method and threshold method to extract complete target,the retention value is set to solve the problem of target void.Then a DPM human target accurate judgment method is designed to combine DPM with ViBe to get more accurate human parts.(2)Motion target tracking.In order to meet the requirement of target detection,an improved target tracking algorithm based on KCF is designed in the case of slight occlusion and change of target proportion.The algorithm uses the similarity between the target area of candidate region and the target area of current frame to judge whether the target is occluded or not,and decides whether to update the target template,so as to solve the problem of slight occlusion of the target.Because of the distance problem,the ratio of detection frames will change,which will affect the extraction of features and lead to tracking failure.Scale change factor is used to automatically change the size of target tracking frames,so as to improve the accuracy of subsequent behavior detection.(3)Behavior detection and classification.In this process,the motion features(such as centroid coordinates,acceleration and minimum outer rectangle)of the target are acquired from the trajectory of the target motion obtained by the tracking algorithm as the basic features of the behavior classification.Fourier descriptor features(such as translation,rotation and scale invariance)are added to these features,and multi-features are fused to form the feature vectors to be detected.SVM classifier is used to train and classify samples.The experiment was carried out by adding falls to KTH public behavior database.The results show that the average classification accuracy(normal walking,running and jogging)of the proposed algorithm is 1.79% higher than that of ViBe+KNN algorithm,so the ViBe+KCF+SVM algorithm can improve the accuracy of behavior detection.The target's fall behavior can also be detected on the increased data of fall behavior.Therefore,the algorithm of behavior detection in this paper can be further applied to the safety alarm processing(call for help and fall)in the daily life of the elderly living alone indoors.