一种自适应的PSO粒子滤波人脸视频跟踪方法

Face Tracking Based on Adaptive PSO Particle Filter

提出了一种自适应的PSO粒子滤波人脸视频跟踪算法。本算法充分利用粒子群算法的寻优能力,使粒子向真实值的后验概率分布移动,同时引入小生境(niche)技术加以改进,构造出多种群特性,使目标分布呈现非线性非高斯特性的多模分布,由此提高对动态系统中最优解动态变化的自适应能力。实验表明,在简单背景匀速运动、复杂背景匀速和变速运动的人脸视频跟踪中,和传统粒子滤波、普通PSO粒子滤波相比,具有良好的跟踪精度和稳定性。

This paper presents a new adaptive PSO particle filter face tracking algorithm. Our algorithm fully utilizes particle swarm optimization（PSO） ability to make the posterior prob- ability distribution movements of the particle, meanwhile introduces the niche technology to improve the particle diversity, then the target distribution is nonlinear non-Gaussian and multi-mode, thus improves dynamic adaptive ability to the optimal solution of the dynamic system. Experimental results show that our algorithm has a good tracking accuracy and sta- bility when comparing with particle filter and traditional PSO in the simple background, complex uniform and variable motion background.