基于均值漂移采样调整和积分直方图表达的粒子滤波跟踪算法

Algorithm Based on Mean Shift Sampling Histogram Adjustment and Integral Expression of Particle Filter Tracking

粒子滤波跟踪算法是对视频目标跟踪常用算法的改进。通过对采样粒子的均值漂移调整,使采样粒子集中于其邻近的局部极大值区域,加快了粒子收敛的速度;通过图像的积分直方图表达方式,对原算法中低效的直方图的统计工作,用粒子所在矩形区域的4个顶点的积分直方图的加减运算代替,运算速度得到较大程度的提高。对实际图像的跟踪和分析表明,本算法和传统的粒子滤波算法相比,具有更快的收敛速度,更短的计算时间,且粒子数越多,粒子所在区域面积越大,本算法的优势越明显。

In this paper,commonly used in video object tracking particle filter tracking algorithm has been improved. By adjusting the sample mean particle drift,the particle could concentrate in its region adjacent local maxima and accelerate the convergence of the particle velocity. This new algorithm could greatly improve the computing speed,by integrating the expression of the image histogram（ the particle is converted to four vertices of the rectangular area where the integral histogram subtraction）,while the original histogram is extremely time- consuming operation algorithm. Compared with the traditional particle filter algorithm,the new tracking and analysis of the actual images show that the algorithm has faster convergence,shorter computation time. And the greater the particle size of the area,this new algorithm has more obvious advantages.