Multi-camera vehicle tracking and re-identification based on visual and spatial-temporal features.

Xiao Tan, Zhigang Wang, Minyue Jiang, Xipeng Yang, Jian Wang, Yuan Gao, Xiangbo Su, Xiaoqing Ye, Yuchen Yuan, Dongliang He, Shilei Wen, Errui Ding

Abstract: Due to the heavy occlusions, large variations in different viewing perspectives and low video resolutions, the tracking and re-identification of vehicles under multi-camera become challenging tasks for the intelligent transportation system (ITS). In this work, we propose a novel framework for multi-camera tracking, which integrates visual features, orientation prediction and temporal-spatial information of the trajectories for optimization. In addition, based on the tracking information generated by our framework, we propose a united method for multi-camera re-identification that takes both visual features and tracking information into account. In order to make the visual feature robust for occlusion and perspective variation, our method adopts various features that are extracted from global image, regions and areas around key points, and the tracking information are also used to refine the retrieval results generated by the visual features. Our algorithm achieves the first place in vehicle re-identification at the NVIDIA AI City Challenge 2019.