Single Object Tracking Model Based on Shot Boundary Detection for Target Re-identification in Classroom Videos

Objects’ mobile behaviors in classroom videos are of great significance to education research. However, the frequent moving in and out of objects, namely the object out-of-view, are to blame for the unsatisfactory tracking results in classroom videos. To address this issue, we proposed a single object tracking model for target re-identification called TeaTrack. Initially, we introduced the Hopcroft Karp algorithm to NeighborTrack to determine if the target is out of the camera. Then, TeaTrack will detect the shot boundary objects through YOLOv5-DeepSORT optimized by alpha-IoU. Eventually, the target will be re-identified by a retrospective recognition module. The experimental results demonstrate that our model outperforms most state-of-the-art methods and can achieve efficient target re-identification.