复杂环境下基于自适应粒子滤波器的目标跟踪

Target Tracking Based on Adaptive Particle Filter Under Complex Background

提出一种基于目标颜色特征的自适应粒子滤波算法,在非遮挡情况下,根据运动预测的准确程度自适应选择粒子数量和运动模型中高斯噪声的方差,保证跟踪的实时性和粒子的有效性;遮挡情况下改变目标的运动模型,使粒子只做布朗运动,并且各粒子经均值漂移算法的一步迭代进行优化,从而可以减少粒子数量以及更快恢复正确的跟踪.实验结果表明该算法具有较强的鲁棒性,能有效实现复杂场景下的目标跟踪.

An adaptive particle filtering algorithm based on color feature is proposed. Under non-occlusion condition, adaptively choose the number of particles and the variance of Gaussian stochastic noise in the dynamic model based on the accuracy degree of the deterministic prediction to guarantee the real-time tracking and valid particles. Under occlusion, the dynamic modal is changed and the particles only do Brownian motion,then the particles are optimized by one-step iteration of the mean shift algorithm tO decrease the number of particles and resume the fight tracking more quickly. Experimental results indicate it is robust and has good performances under complex background.