Improved YOLOv5 Algorithm for Intensive Pedestrian Detection

Pedestrian detection is an important basic research topic in the field of target detection, which can provide effective information support for public places with large flow density such as shopping malls and scenic spots as well as intelligent security fields. To solve the problem of tracking target loss and low detection and recognition rate caused by pedestrian occlusion or portrait over-lap in pedestrian target detection in a crowded area or scene, an improved YOLOv5 algorithm integrating the attention mechanism was proposed. The relationship between feature channels and spatial information of the feature map was deeply mined by introducing an attention mechanism to further enhance feature extraction of pedestrian target visual area. To improve the feature fusion ability, Bidirectional Feature Pyramid Network (BiFPN) feature pyramid was used to enrich the cross-scale connection mode, preserve the shallow layer characteristics, and improve the detection accuracy. To improve the convergence ability of the model, EIoU was used to replace the original loss function of YOLOv5 to optimize the regression prediction of the anchor, which reduces the training difficulty of the network and improves the detection rate under occlusion. Compared with the general YOLOv5 algorithm, the improved algorithm proposed in this paper has higher accuracy and a lower missing rate in pedestrian detection in crowded areas or scenes, while the real-time performance of the original algorithm is still maintained.