A systematic framework for real-time online multi-object tracking

Multi-object tracking has been studied extensively and is still challenging in the computer vision research field because there is uncertainty about targets due to similarity of targets, complex target interactions, occlusions in the long period, and cluttered, dynamic backgrounds. However, there are very rare attempts to realize multi-object tracking by analyzing the requirements, breaking it down into sub-problems, and designing implementations for phased development and verification. To this end, we propose a systemic multi-object tracking framework for solving the sub-problems by analyzing the causes of performance degradation and defining difficulties of implementation and verification. Moreover, we implemented the systemic framework for online multi-object tracking and verified its performance.