Pedestrian Detection and Feedback Application Based on YOLOv5s and DeepSORT

To solve the problem of low counting efficiency of people flow supervision in scenic spots and cities, an improved YOLOv5s detection method combined with DeepSORT target tracking is proposed. As the accuracy of DeepSORT multi-target tracking depends on the detection efficiency of the target detection algorithm, the attention module CBAM is integrated with the Neck part of YOLOv5s network to improve the ability of the target detection model to extract small object features, thereby improving the recognition ability. The Market-1501 data set is used to train the pedestrian re-id model, and the images containing pedestrians in the VOC data set are used as the training set to train the detection model so that the model only detects pedestrians. Finally, connect the improved YOLOv5s detector and DeepSORT, and set a virtual detection line in the video to count the flow of people. The experimental results show that the average accuracy of the improved YOLOv5s is 1% higher than that of the original algorithm, and combined with DeepSORT tracking, the detection rate of 34 frames is achieved in the test video.