以视频目标跟踪中粒子滤波的粒子采样优化设计为研究内容,提出一种自适应分层采样辅助粒子滤波算法,以实现保证跟踪准确度和兼顾跟踪鲁棒性的要求.以Bhattacharyya系数为参量设计了粒子数调节函数,能够根据跟踪质量在粒子集中自适应分配用于保证准确度的粒子数和维持鲁棒性的粒子数.以最小二乘法对目标运动的预测点作为产生新粒子集的均值偏移操作起点,使新粒子集更准确的描述目标似然分布并提高算法效率.不同场景下的跟踪实验表明,算法能很好的应用于遮挡和运动方向渐变等情况下的跟踪,处理时间满足实时性要求.

For the particle′s sampling and optimization of particle filter in video object tracking,an adaptive layered-sampling auxiliary particle filter is proposed to realize the tracking precision and robustness′s requirement.A function which adjusts the particle amount is designed based on the Baattacharyya coefficient.According to the tracking quantity,it can allocate the particles keeping precision and the particles keeping robustness adaptively in the particle set.A predicted point of object-moving based on the LSM (least square method) algorithm is used to be the origin to do the meanshift operation.By this operation,the new particle set is made,it will be better to describe the object′s likelihood distribution and improve the algorithm efficiency.The tracking experiments under different enviroments demonstrate that this algorithm can achieve better tracking performance in presence of occlusion and object-moving in slow and variable direction,the consuming time is up to the requiment of real-time performance.