基于稀疏性理论与子问题耦合的多目标跟踪方法

Multi-target Tracking Method Based on Sparse Theory and Sub-problem Coupling

针对多数目标跟踪方法在非受控环境中稳定性不高以及检测-跟踪模块分离的缺点,提出一种稀疏性检测器与网络数据关联技术相结合的多目标跟踪方法。离散化目标的移动空间,对于3D的每个可能位置,将目标投影到图像平面,形成码字并构建字典。扩展模型至多类别跟踪情况,并根据耦合公式分配给子问题和协调局部解以实现解的最优化。使用网络单纯形算法解决最小成本流数据关联问题。在BU-Marathon,PETS2009等公开数据集上的实验结果表明,与能处理遮挡的多目标跟踪方法相比,该方法具有较高的跟踪精度,误检率和漏检率更低。

Aiming at the inherent shortcomings that many existing target tracking methods lack of stability in non controlled environment and the detecting-tracking module are separate, a multi-target tracking method based on sparseness detector and network data association technique is proposed. The moving space of the target is discretized, and the target is projected onto the image plane for each possible 3D position. The dictionary is built after formation of code. The model is extended to multiple classes. The optimal solution is achieved by assigning coupling equation to the sub problems and coordinating local solutions. The minimum cost flow problem of data association is solved by network simplex algorithm. BU-Marathon, PETS2009 and other public data sets are used in the experiments. Experimental results show that, compared with the multiple target tracking algorithms with occlusion handling, the proposed method has higher tracking accuracy and less false detection rate and missing detection rate.