BGM-SA: Simulated Annealing Bipartite Graph Matching Algorithm Based on Visual Prior Knowledge

Realizing fast and accurate object detection and tracking models is one of the research hotspots in the field of computer vision. However, with the continuous growth in demand for detecting and tracking in crowded scenes and small objects, the scale of bipartite graph matching between prediction boxes and target boxes becomes increasingly larger, significantly affecting the real-time performance of algorithm models. Therefore, this paper proposes a bipartite graph matching optimization algorithm based on simulated annealing. This algorithm integrates prior knowledge from the computer vision field and improves upon traditional simulated annealing methods, greatly accelerating the resolution of bipartite graph matching problems in computer vision, particularly suitable for large-scale scene scenarios. Experimental results demonstrate that compared to the improved KM algorithm in the SciPy library as a benchmark, the proposed algorithm achieves higher speed and lower error rates in large-scale scenes.