基于目标检测和图割的多目标跟踪算法

多目标跟踪是智能视频监控中的一个具有挑战性的问题.提出了一种基于运动目标检测和图割理论的多目标跟踪算法.首先,利用码本模型对背景建模,检测运动目标.然后,令一个标签对应一个目标,建立能量方程,把多目标跟踪问题转化为能量最小化的组合优化问题.最后,构造网络图,利用最大流-最小割算法寻找最优解.实验结果表明,所提算法能够处理新目标的进入和原有目标的离开,对多目标之间的遮挡具有较强的鲁棒性.

Multi-target tracking algorithm based on object detection and graph cuts

Tracking of multiple targets is a challenging problem in intelligence video surveillance. A novel multi-target tracking algorithm based on moving objects detection and graph cuts is proposed. Firstly, moving objects are detected by modeling the background using codebook model. Then, the energy function is established, making each label correspond to one object. The problem of multi-target tracking is transformed into a combinatorial optimization of energy minimization. Finally, a network is constructed and the optimized result is gained by the max-flow min-cuts algorithm. Experimental results show that the proposed algorithm can deal with the entrance of new object and the exit of existing one in the scene, and it has nice robustness to the occlusion between multiple objects.