【作者】 李琳嶙； 费树岷； 金立左；

【Author】 LI Lin-lin,FEI Shu-min,JIN Li-zuo (Department of Automation,Southeast University,Nanjing Jiangsu 210096,China)

【机构】 东南大学自动化学院；

【摘要】 传统Mean-Shift目标跟踪算法对背景因素比较敏感,对尺寸逐渐变大的目标不能实现稳定的跟踪。本文采用最大后验概率判别指标作为图像匹配的相似性度量,当目标尺寸逐渐变大时通过后向跟踪对跟踪目标中心位置补偿配准,最后使用多尺度图像空间信息量度量的方法,自适应地调整目标跟踪窗口尺寸的大小。实验结果表明,该方法实现了在复杂环境背景下对尺寸变化的运动目标更加有效的、准确的跟踪。更多还原

【Abstract】 Traditional Mean-Shift object tracking algorithm is rather sensitive to the background environment and fails to track the target changing in a larger size.This paper uses an image matching similarity criterion based on maximum posterior probability measure,and backward tracking is used for object centroid registration when the size of the object changing.The information measure of multi-scale image in scale space is used to adaptively adjust the window size of the tracking object.Experimental results show that this method can effectively and accurately track moving object changing in size in the cluttering background.更多还原