目标跟踪中参考模型的自动初始化方法

目标跟踪中参考模型的初始化大多需要手动完成,现有的自动初始化方法也存在明显缺陷,针对这些问题提出了一种自动、精确初始化参考模型的方法。首先采用连通组分标记算法对目标进行计数,并计算每个目标的位置;然后根据回转半径张量法求出包含目标的最小外接矩形;最后从最小外接矩形的参数出发用蒙特卡洛算法求出参考模型的估计。实验结果表明在不同的场景下算法均能准确完成参考模型的初始化,计算出的模型能够体现目标的视觉特征。

An Automatic Initialization Method of Reference Model in Target Tracking

The initialization of reference model in target tracking is usually performed manually; and there are still de- fects in current automatic method. To resolve these problems, this paper proposes an automatically and accurately ini- tialization method. Firstly, the targets number and position are computed using connected components labeling algo- rithm. Consequently, minimal circumscribed rectangles of the targets are obtained according to radius of gyration tensot method. Finally, the reference models are estimated adopting Monte Carlo simulation. Experimental results show that the reference models can be accurately computed in different scenes; also they can reflect the visual features of the targets.