摘要
针对实时图像跟踪中目标尺度不断变化的问题,提出了一种新的最大后验概率指标下尺度自适应的多分辨图像跟踪算法.首先证明了后验概率指标的像素级计算特性,在该特性的基础上提出了一种最大后验跟踪算法.由于后验概率指标不仅可以按照特征进行计算,还可以按照像素进行计算,从而可以方便地实现不同尺度上的像素相似度贡献值的计算和比较,据此提出了一种新的目标尺寸自适应算法.此外,当目标尺寸较大时,可以采用不同的分辨率来计算候选匹配区域的匹配概率值,大大降低计算量,从而保证实时跟踪的时间需求.综合上述特点,给出了最大后验概率指标下目标尺寸自适应的多分辨图像跟踪算法.多组视频跟踪实验结果表明了本算法的有效性.
In order to solve the target scaling problem in visual tracking, a new adaptive multi- resolutional image tracking algorithm employing posteriori probability similarity measure is developed. The pixel-wise computational property of the posteriori probability measure is proved firstly, based on which, a maximum posteriori probability tracking algorithm is developed. The posteriori probability measure can be computed both by feature and by pixel, which enables the convenient computation of each pixel＇s contribution to the similarity value. Based on this property, a new adaptive scaling method is developed. On the other hand, when the size of the tracked target becomes large, multi-resolutional skill can be employed to reduce the computation burden, which is also derived from the computation property of the employed measure. Finally, a new adaptive multi- resolutional image tracking algorithm based on posteriori probability similarity measure is constructed. Experimental results demonstrate the effectiveness of the new algorithm.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2012年第8期1708-1714,共7页
Journal of Computer Research and Development
基金
国家自然科学基金项目(61105034
60905044)
高等学校博士学科点专项科研基金项目(20100201120040)
中国博士后科学基金项目(20110491662)
关键词
图像跟踪
尺度自适应
多分辨
后验概率指标
相似度指标
image tracking
adaptive scaling
multi-resolution
posteriori probability measure
similarity measure
