摘要
Mean shift算法是一种重要的目标跟踪方法。本文在充分研究Mean shift算法的基础上,发现传统的跟踪算法缺乏描述目标的信息,且易受到光照变化等因素的影响,可能直接导致目标跟踪的失败或者带来较大的误差。本文首先计算目标图像区域中基于局部二值模式(LBP)的纹理特征模型,然后在建模目标模型和候选目标模型时采用一种加权背景直方图的方法。该方法能充分考虑目标附近的背景干扰以及背景信息,能有效降低背景和目标位置的相关性。实验表明,改进的方法能有效提高目标的跟踪精度。
Mean shift is a useful tracking algorithm in computer vision. In image sequence, due to the sensitivity to illumination variation and the lack of information for object representation, object tracking will have a failure or a big error tracking. A novel tracking algorithm is proposed in this paper, we firstly compute texture model based on the local binary pattern in object＇s region, and then use a background-weighted histogram method for building object model and candidate object model, it can effectively reduce background＇s interference in target localization. The experimental results show that this improved algorithm can effectively improve tracking accuracy.
出处
《科技通报》
北大核心
2013年第12期169-171,共3页
Bulletin of Science and Technology
基金
新世纪广西高等教育教学改革工程课题(2010JGA185)
