Elderly Helper Detection based on R2-yolov5

Abstract

In order to assess the independent walking ability of elderly people, a system for assessing the independent walking ability of elderly people using yolov5 target detection was developed, and the accuracy of the detection was improved with a modified R2-yo1ov based on yolov5 by replacing the residual structure bottleneck in its unique C3 structure with the Res2net residual module, and by channel-wise and activation function optimization in terms of channels and activation functions. In order to enhance the information transfer between network layers, the upper and lower feature layers are fused to improve the detection effect. The experimental results show that the map of R2-yo1ov5 tested on the elderly helper dataset can reach 96.7%, which is 1.8% higher than the original yolov5 network, and the detection effect of support class is improved by 5.5%, which is a significant improvement in the detection effect and can meet the requirements of the detection scenario.