Abstract: Aiming at the problem of low re-identification rate caused by changes in internal and external conditions such as pose, external occlusion, illumination and camera parameters, this paper presents a video person re-identification algorithm based on spatio-temporal feature and camera-correlation feature augmentation. Firstly, the spatialtemporal gradient histogram (HOG3D) based on video and the apparent features based on the images are extracted as the feature descriptor; and then the distance metric is combined with the Marginal Fisher Analysis (MFA) algorithm to augment the feature, with the purpose that enhancing the connection between view-generic feature. Finally, the Euclidean distance measurement and sorting are performed. In addition, two experiments are carried out in the two video person data sets of iLIDS-VID and PRID 2011. Experiments show that this algorithm can make full use of the motion information contained in the video to obtain a robust video person re-id model, and the person re-identification matching rate has been effectively improved.