目标颜色特征模型.docVIP

基于联合特征的边缘粒子滤波目标跟踪算法研究* 孟军英1，刘教民2，韩 明2 (1.石家庄学院 计算机科学与工程学院,石家庄 050035；2. 燕山大学 信息科学与工程学院,河北 秦皇岛 066004) 摘 要: 针对采用单一特征建立的动态空间模型与实际系统差距较大，从而使估计误差增加的问题，通过将系统的状态参数引入颜色特征模型中，与颜色特征参数一起构成系统状态空间向量，提出了一种联合颜色状态特征的优化目标跟踪算法。应用Rao-Blackwellization算法思想，由Kalman线性滤波方法解析的处理线性的颜色特征转移和更新过程；而目标位置参数采用粒子滤波进行估计，提高了视频目标跟踪的精度和实时性。通过与其他相似算法的比较实验结果可以看出，算法在环境亮度发生变化、目标遮挡等情况下，仍能够保持较高的跟踪精度，既提高了跟踪系统的鲁棒性，又保证了算法的实时性，优于传统的单一特征视频跟踪算法。 关键词: 目标跟踪; 边缘粒子滤波；核概率密度估计；卡尔曼滤波 中图分类号: TP391.4　　　 文献标志码: A Marginalized particle filter for combined feature target-tracking MENG Jun-ying1, LIU Jiao-min2, HAN Ming2 (1. Dept. of Computer science, Shijiazhuang University, Shijiazhuang Hebei 050035, China；2. College of Information Science and Engineering, Yanshan University, Qinhuangdao Hebei 066004, China) Abstract: In order to improve accuracy of system model and reduce evaluated error resulting, this paper proposed a marginalized particle filter algorithm with joint characteristic. Depart from color appearance parameters, the location parameter introduced into system model as another system state vector. Furthermore, split the state into two parts: the location parameters, which are estimated stochastically by particle filter, and the color appearance parameters, which are estimated analytically. Both parts are jointly tracked using Rao–blackwellization technology. Rao-blackwellization algorithm reduced the dimension of mixed linear/nonlinear system and realizing a higher tracking precision and lower computation complexity. Comparison experiment results demonstrated that the proposed algorithm is able to automatically update when ambient brightness and speed of target changes or the target is sheltered. The algorithm is more robustness and real-timeness than the traditional color-based particle filter tracking algorithm. Key words: target-tracking; marginalized particle filter; kernel density estimation; Kalman filtering algorithm 引言 在视频监控系统中，跟踪移动的目标物体是一个基本的功能。目前，已经有许多跟踪算法应用于监控系统中。其中，