基于目标运动模型的跟踪方法

Tracking Algorithm Based on Target Motion Model

针对常用的机动目标模型不能准确描述目标实际运动规律及常用跟踪算法只拟合目标的形心而不是目标轮廓轨迹的问题,提出一种基于目标运动模型的跟踪算法。该算法提取已检测出目标轮廓上的角点作为样本点,采用神经网络来构建目标运动模型,将用此模型预测出的目标轮廓上的点作为主动轮廓线的初始控制点来检测出目标真实轮廓,并反馈回神经网络的输入端来修正模型误差。实验结果表明该跟踪算法能很好地将前续目标检测结果继承到后续的目标检测过程中,对于目标跟踪中的遮挡问题也能很好地解决。

A tracking algorithm based on target motion model was proposed since the actual motion rule of target could not be accurately expressed by common mobile target model and only the centroid trace was predicted by tracking filter. The method extracts the corners of the target contour as sample points of BP neural network, and then gets the motion model of the target after training the BP neural network. Taking points predicted by BP neural network as initial controlling points of active contour, the real target contour could be detected, which are fed back to BP neural network to modify the model error. The experimental results show that previous target detective results are succeeded to the subsequent target detection process by the tracking algorithm and the occlusion problem of the target is well solved by the algorithm.