Old Man Fall Detection Based On Surveillance Video

The number of elderly people in our country is increasing,and the proportion of elderly people is getting higher and higher,declaring that our aging society has arrived.The older the elderly,the greater the likelihood of a fall.Once the old man falls,it will cause serious injury to the old man and bring a heavy financial burden to the old man's family.In recent years,computer vision has developed rapidly,and target detection algorithms have also achieved good results.In order to reduce the harm caused by the fall of the elderly,it is of great practical significance to apply the target detection technology to the fall of the elderly.It can be found immediately after the fall of the elderly,so as to minimize the harm of the fall to the elderly.Through in-depth research on algorithms in the field of target detection,the Deep-SORT and YOLOv3 algorithms are used to realize real-time fall detection of human targets in surveillance video scenes.Experimental results prove that the combination of YOLOv3 algorithm and Deep-SORT algorithm can detect the fall of the elderly.This article mainly did the following work.1.Make a data set: Divide the human body posture into fall and non-fall,collect the fall data and mark it as a fall data set;use the camera to automatically take pictures and download the network to obtain the face data set.Second,the training data set: first use the K-means++ algorithm to cluster to obtain 9 anchor values,and then use YOLOv3 to train the fall data set to obtain the weight file,and select the best weight file through the test.Third,target tracking:combining the weight file provided by YOLO's official website with Deep-SORT to realize human multi-target tracking and cutting the tracked target image.4.Face recognition: First collect the required face data set,then train the data set to obtain a face recognition model,and finally perform face recognition on the tracked target.5.Fall detection: Use the weight file obtained by training to detect the target image,determine the occurrence of a fall event according to certain rules and report to the police.In addition,this article tunes the relevant parameters of YOLOv3 and Deep-SORT to make it more in line with the needs of the fall detection system,and finally shows a good fall detection effect.