Indoor Fall Detection Based on Yolov5 And Openpose

The current traditional fall behavior detection method constructs a fall detection model by pose annotation of the acquired images, which leads to low detection accuracy due to the lack of lightweight processing of the model. In this regard, indoor fall detection based on Yolov5 and openpose is proposed. The Yolov5 network structure is constructed and the path aggregation network is added on top of it. The skeletal joint points of human body are extracted using the openpose algorithm, and the fall behavior is determined based on the skeletal joint point information obtained in real time, and finally the model is lightened. In the experiment, the proposed model is verified for detection accuracy. The analysis of the experimental results shows that when the proposed model is used for indoor fall behavior detection, the MAP value of the model is high and has a high inspection accuracy.