Multi-Object Tracking Algorithm for Detecting Entry and Exit Events in Indoor Surveillance
Yang‐Yang Chen, Xiaohan Hu · 2023
In indoor surveillance scenarios, there is a lack of intelligence in the analysis and processing of surveillance video data, and there is a lack of simple and effective methods for personnel entry and exit detection management. In response to these problems, this paper proposes a pedestrian entry and exit event detection algorithm based on multi-objective tracking in indoor surveillance scenarios, named Entry and Exist Judger (EAEJ). It simply requires monitoring video data to achieve intelligent analysis in monitoring scenarios, detecting personnel entry and exit behavior and item carrying status. In order to eliminate the impact of occlusion on the accuracy of multi target tracking, a multi target tracking model framework DKTrack considering depth information and appearance features is proposed. In response to the lack of public datasets in indoor monitoring scenarios, images and videos were collected and captured, and a dataset called DoorKeeper was constructed to fill the gap in the dataset for this task. DKTrack conducted performance validation on MOT17 and MOT20, and the results showed that the algorithm improvement brought about performance improvement. EAEJ has achieved extremely high accuracy in DoorKeeper for entry and exit event detection tasks.