一种基于纹理和颜色的目标跟踪方法

摘要

研究视频物体识别系统,传统连续自适应均值偏移(Camshift)跟踪方法根据H分量建立被跟踪目标的颜色模型,而H分量易受亮度(V分量)的影响,造成不能准确跟踪运动目标.为解决上述问题,引入运动目标的纹理特征,先提取HSV颜色空间的H分量,把它转化为局部二元纹理(LBP)图,计算目标的LBP纹理直方图并把反向投影到LBP纹理图上,得到LBP纹理概率图,然后采用Camshift算法确定当前图像中目标的尺寸和中心位置.对手势和人脸跟踪进行仿真计算,实验结果表明,在跟踪过程中可以对目标进行稳定的实时跟踪,通过计算,也改善了传统方法,使识别人脸不受光照的影响,验证了改进方法的有效性.%Traditional Camshift is a tracking algorithm based on H channel and can be disturbed by the illumination easily. The paper proposes a tracking algorithm based on Local Binary Pattern (LBP) feature which is combined with color feature. In the process of object tracking, H channel is separated from the HSV space and is converted to LBP image at first. Then Camshift operates on a back-projection image produced from object histogram model. The experiment result shows that the tracking algorithm proposed can track object stably and in real-time, regardless of the existence of coveting, changed appearance and illumination changing.