Dense Pedestrian Detection Method Based on Improved YOLOv5

For the current increasing number of pedestrian traffic accidents, driving vehicles can accurately and quickly detect pedestrians is one of the effective ways to avoid accidents. Aiming at the phenomena of omission and false detection of pedestrian detection in dense places by the current pedestrian detection model, a method is proposed to replace the CSP2_1 module with the BoT3 module based on the self-attention mechanism in the backbone network to improve the model's ability of extracting the global features; and to add the lightweight Hybrid Attention Mechanism HAM (Hybrid Attention Module) in the output side of the backbone network to enhance the model's ability of capturing important features, accelerate the convergence speed, and improve the model's convergence efficiency. Enhance the model's ability to capture important features to improve the model convergence efficiency, accelerate the convergence speed and improve the accuracy of the model. The experimental results on the homemade pedestrian detection dataset show that the average accuracy of the improved YOLOv5 model reaches 90.9%, which is 2.1% higher than that of YOLOv5. With little effect on the detection speed, the pedestrian missed detection is reduced and the detection accuracy is improved.