基于最小费用流模型的无重叠视域多摄像机目标关联算法

A Min-cost Flow Based Algorithm for Objects Association of Multiple Non-overlapping Cameras

二分图最大匹配算法是常用的无重叠视域多摄像机目标关联算法,本文提出了一种基于最小费用流模型的关联算法,并与前者进行对比.实验发现前者很大程度上依赖于效用函数的定义,效用函数存在的偏差导致该算法求解结果不理想.后者理论上能够估算并修正效用函数的偏差,得到更优的解.本文进行了大量仿真实验,实验表明了本文算法更为鲁棒有效.

Maximum matching in bipartite graph is always used in objects association of multiple non-overlapping cameras. This paper proposes a new association algorithm based on minimum cost flow. Through experiments, we find that the maximum matching algorithm largely depends on the utility function, which may affect the results. The proposed algorithm can correct the errors of the utility function and then obtain better results. Simulation results show that the proposed algorithm obtains better results and is more robust.