A Method for Multi-view Human Re-identification in Substations

Re-identification(re-ID) is matching or identifying the same target across viewpoints. In substation monitoring, it is crucial to observe staff to prevent irregular activities. Due to frequent occlusions, a multi-camera system is necessary. Thus, we use re- ID technology to track people. Considering severe occlusions, uniform appearance and temporal information, we propose a method that extracts appearance, skeleton and temporal features. We introduce Temporal Memory Module (TMM) to obtain temporal information. To avoid addressing particular memory keys continually, we use temporal memory spread loss (Ls<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">T</inf>) as solution. We treat keypoints of skeleton as nodes of graph to achieve higher-order information. Then alignment can be treated as graph matching to improve accuracy. The experiment of re- ID on the data collected in the substation has proved effectiveness of our model.