基于粒子滤波的彩色图像跟踪  

Color image tracking algorithm based on particle filter

作　　者：吴川[1] 杨冬[2] 郝志成[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033 [2]长春工业大学计算机科学与工程学院,吉林长春130012

出　　处：《光学精密工程》2009年第10期2542-2547,共6页Optics and Precision Engineering

基　　金：中国科学院青年基金资助项目(No.O76031T070)

摘　　要：视频跟踪中的目标检测和目标跟踪通常不能通过一个算法同时完成,而是需要两个计算法则,过程复杂,耗时较多。为了实现序列彩色图像的实时检测与跟踪,本文提出了一种基于粒子滤波的实时目标跟踪算法。以目标有无信息和目标位置信息为变量建立了联合状态向量,利用粒子滤波方法实现目标检测及跟踪。为了减少计算量,在充分考虑跟踪区域各像素权重的条件下,建立基于颜色信息的特征直方图作为观测向量,并用于后验估计。实验表明,本文提出的方法在选择150个粒子的情况下,对于768 pixel×576 pixel大小的彩色图像,能够在14.37 ms内检测并跟踪目标,而且能在目标发生旋转变化和尺度变化时,保持稳定跟踪,证明了该方法具有一定的鲁棒性。In video tracking, target detection and target tracking can not be implemented by an algo- rithm synchronously, so the image processing is complex and processing time is very long. To imple- ment the target detection and target tracking of sequential color images in real time, a target tracking algorithm was presented based on a particle filtering in this paper. By taking the target information and target positions as the variables, a hybrid valued sequential state vector is established, and the target detection and the target tracking are implemented synchronously by using the particle filtering method. Moreover, in order to reduce process time, the color-based feature histogram is presented based on the pixel position weight in tracking fields as an observing vector, and then it is usedin the posterior estimation successfully. Experimental results indicate that a target can be detected and tracked synchronously in ]4. 37 ms by using 150 particles for a color image with a size of 768 pixel× 576 pixel. Obtained results also prove that the method has higher robustness due to its stable tracking state in the rotation and scalar change of the target.

关 键 词：目标检测 目标跟踪 粒子滤波 特征直方图 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TN941.1[自动化与计算机技术—计算机科学与技术]