Data association method based on scenario analysis using weak tracking data for multi-target tracking

摘要

Video surveillance has been used extensively in our daily life today, for example, traffic safety, trajectory analysis, event monitoring, and crowd behavior identification. However, sometimes in many real-world video sequences dealing with complex scenario and environmental noise is still a challenging task. One solution is to consider information of scenario during tracking. In this paper, we propose scenario analysis using weak tracking data to detect the entry and exist regions in the scene. The regions are learnt according the historical frame. Weak tracking data is short and maybe broken trajectory. This scene information is then used in data association process of object tracking. Finally, the scene information with a modified data association to tracking multi-targets. The experiments indicates our methods are usefulness and reliability.