Research On Human Fall Detection Algorithm Based On Video In Home Environment

At present,the aging of the world population is becoming more and more serious.Falls are the main cause of death and injury of the elderly over 65 years old,and fa ll events have a serious impact on the life and health of the elderly in China.Therefo re,it is of great significance and value to study fall detection.In this paper,a fall det ection algorithm based on support vector machine and threshold judgment is proposed based on computer vision,The Fall detection algorithm was tested and compared with the public Dataset UR Fall Dataset and the sample data of self-shot Dataset.The expe rimental results show that the proposed algorithm has a good detection effect.Firstly,this paper uses YOLOv3 target detection algorithm to replace the traditiona l target detection method.By training sample image data to accurately detect the positi on of the human body and the head of the human body in the video,The YOLOv3 a lgorithm was compared with SSD algorithm and Faster R-CNN algorithm through expe riments.The results show that it is better than the other two algorithms in the accurac y and real time of target detection.The combination of YOLOn3 algorithm and Deeps ort algorithm is proposed to track the detected target.The combination of detection an d tracking is applied to the field of fall detection.Through experiments,it is verified t hat the fusion of the two algorithms has better tracking effect on the detected target t han that of the single Deepsort algorithm.Secondly,in order to well represent the behavioral characteristics of the human body,this paper extracted the characteristics of the aspect ratio of the human body,the rate of change of the central position,the eccentricity of the fitting ellipse and the rate of change of the position of the human head,and analyzed the six most common forms of the human body: walking,squatting,sitting down,bending down,lying down and falling down have been tested.The results show that the characteristics selected in this paper can well represent the behavior state of human body.Finally,this paper proposes a threshold judgment combined with support vector machine based fall detection algorithm,to detect target position detection and tracking,feature extraction,based on the obtained from extract the features of the first to use screening part of eligible image threshold judgment,after the application of support vector machine classification,so as to determine whether the target fall,And compared with the algorithm using convolutional neural network to directly judge whether a human falls,the comparison results show that the algorithms proposed in this paper are able to judge whether a human falls well,and meet the real-time requirements.To sum up,the fall detection algorithm based on support vector machine combined with threshold judgment proposed in this paper has a very good detection effect in the public Dataset UR Fall Dataset and the sample data of self-shot Dataset.