Multiple Group Detection and Tracking for Multiple Domain Formation

Addressing the problem of multiple group detection and tracking from multiple domain formation, a universal Joint Multiple Group Detection and Tracking (JMGDT) method is proposed. Firstly, JMGDT uses an adaptive parameter estimation algorithm for threshold estimation which improves the group detection accuracy. Additionally, based on the proposed threshold estimation and directed graph transitive closure, the method could discriminate the subjected relations between objects and groups. Furthermore, JMGDT uses a bilayer graph expression for describing the relations between groups which is used for group similarity measurement based on attention mechanism. Through comparative experiments on both real and simulated datasets, the proposed method has demonstrated certain advantages in several key metrics. The experimental results show that the proposed method in this paper has achieved an average improvement of about 3% in Multiple Object Tracking Accuracy(MOTA ) and about 5% in Group Detection Success Rate(GDSR), effectively solving issues such as insufficient group detection accuracy, ambiguous individual discrimination, and complex similarity calculation.