Research of Pedestrian Re-identification Method Based on Video Surveillance

In order to solve the problem of person recognition in cross-view video sequences of non-overlapping camera, most of the current person re-identification models based on deep learning either need to manually label features as their attributes, or learn the overall single semantic level of feature representation. This paper proposes a person re-identification method based on DNN with multi-level feature fusion, it can automatically learn multi-level discriminative visual factors that are insensitive to viewing condition changes, and identify and utilize them when matching images. Firstly, this paper uses the HOG feature to perform person detection on the video of the two cameras respectively. The person images detected of the camera1 are used as the prob, the person images detected in the camera2 are used as the gallery, and then the two parts are put into the person re-ID model and completed by the training. Finally, the cross-view tracking is implemented for the re-identified persons in combination with the KCF algorithm. The experimental results confirm the accuracy and efficiency of the method.