自动选取特征的行人跟踪

根据非刚性物体运动的特点，提出一种利用支持向量机有效对多目标进行分离并能自动选择目标最明显特性区域的覆盖比算法，克服了以往对选取区域事先进行设定的情形。采用目前常用的加权颜色直方图作为备选粒子，进行行人跟踪。由于建立了合理目标颜色模型，减少了对粒子数量的需求，降低了计算复杂度，有利于实现实时跟踪。利用Bhattacharyya距离描述粒子与目标颜色模型的相似性，作为粒子更新权值的依据。试验结果证明了自动选取被跟踪目标的特征与事先设定备选区域有等价效果。

Pedestrian tracking with automatic selection of characteristics

According to the non-rigid characteristics of the moving target, an algorithm based on Mantle ratio, which can effectively separate multi-target using Support Vector Machine (SVM) and automatically select the largest characteristic region of non-rigid was proposed. It canceled the special requirements of selected regions beforehand. And then the general method of regarding the weighted color histogram as a particle and track pedestrian was adopted, which can reduce the requirement on the number of particles and the computational complexity, so it is helpful to realize the real-time track. Furthermore, the Bhattacharyya distance was employed to estimate the similarity between the color model and the particles, and the distance value was the base of particle update. The experimental results prove the equivalance between automatic selection of characteristics of tracked target and setting the selected regions.