Search

Aiming at the problems of similar appearance and occlusion of people in the video person re-identification, a video-based person re-identification model based on pyramid segmentation and attention mechanism was studied and designed. First, in order to enhance the recognition ability of the graph model for the local features of pedestrians, a multi-scale horizontal pyramid segmentation method was proposed. In addition, given that the simple spatiotemporal attention module was prone to damage person features due to occlusion, the spatiotemporal attention module was improved using the spatiotemporal correlation attention method, which gradually learns and aggregates spatially local information while interacting in time sequence to suppress person interference features and enhance discriminative features. This paper evaluates the model on Mars and DukeMTMC-VideoReID datasets, and the experimental results confirm the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]