Human Body Tracking Method Based on Deep Learning Object Detection

Aiming at the problem of poor robustness of human detector based on artificial extraction feature, This paper proposes a visual tracking method based on deep learning object detection, which draws on the research results of target detection. The method utilizes the advantage of deep learning in feature representation, and uses the regression-based depth detection model YOLO to extract candidate targets. We re-clustered the data set for human targets, which improved the network performance of YOLO. For the extracted candidate frame position, the region is clipped. The HOG features of the candidate regions are extracted for target screening to achieve target tracking. Compared with pedestrian detection methods such as KCF and so on, this method reduces the miss detection rate and false detection rate, improves the robustness of the algorithm, and the detection speed meets the real-time requirements.