行人跟踪的多特征融合算法研究

The Research for Pedestrian Tracking Algorithm with Mutil-feature Fusion

为了克服遮挡,准确跟踪目标,本论文提出了一种基于最邻近法的多特征混合的跟踪算法。颜色特征和几何特征是视觉跟踪中最直观的特征,而且这两种特征的提取和匹配用时较少,被跟踪目标在没有发生遮挡时,使用颜色特征和几何特征也能准确跟踪被跟踪目标。当发生遮挡时,只依靠被跟踪目标的颜色特征和几何特征将不能继续跟踪。Harris角点可以应对部分遮挡,所以将这三种特征融合起来就能很好的克服遮挡问题。但多特征融合往往会降低系统的时效性,本文采用最邻近法来决定目标匹配的优先度,克服了多特征对系统实时性的影响。实验结果表明,本文提出的算法对目标形变及遮挡具有良好的跟踪准确性和鲁棒性,并且克服了特征融合带来的时效性差的问题。

In order to overcome occlusion and to accurately track objects, the thesis proposed a kind of multi-featured tracking algorithm, which is based on the Nearest Neighbor. The color and geometrical features are the most visualized and they can be used to track objects accurately. When the occlusion appeared, according to the color and geometrical features of the target object will be unable to track the target. Harris comer can be utilized to deal with some occlusion. Therefore, these three features, namely color, geometry and Harris, can be mixed together to overcome occlusion. However, the mix- ture of many features will reduce the timeliness of the system. The thesis decided the priority of the object matching by u- sing nearest neighbor and overcome the bad effects of multi-features on timeliness. As experimental results show, the algo- rithm proposed in this paper has good tracking accuracy and robustness to the target＇ s deformation and block, and overcome the problem of poor timeliness which bring with feature fusion.