A Pedestrian Detection Method Based on Improved YOLOv5s

Abstract

Aiming at the lack of accuracy and effect in some scenarios when the YOLOv5s algorithm is applied to pedestrian detection tasks, we designed the InECA_YOLOv5s algorithm to improve the accuracy of the original algorithm and enhance the recognition effect of the algorithm in some special scenarios. First, a small object detection head is added to the original model, and the implementation of some activation functions and convolution units is modified; then, an efficient channel attention module is inserted at the appropriate position in the model; Finally, the calculation strategy of the intersection and union ratio of the network in the training phase is adjusted. Experiments on the self-made multi-scene pedestrian detection dataset show that the improved algorithm effectively improves the accuracy.