摘要
Mean-Shift算法是一种简单高效的目标识别算法,但是不能有效地识别被遮挡的目标和有尺度变化的目标。基于仿射变换,提出了一种尺度自适应的机器人目标跟踪算法。定义了转角点,并根据转角点匹配对目标进行区分,最后通过仿射变换识别出目标的尺度变化。与其它相关算法相比,该算法能有效地识别被跟踪目标的遮挡问题;当被跟踪目标的尺度发生改变时,该算法仍然能准确地对目标进行识别。分析表明,当视屏流中每秒的图像小于25帧并且目标的图像小于2×104个像素时,该算法可以用于目标的实时跟踪。
Mean-Shift algorithm is a simple and efficient target tracking algorithm,but it can't recognize occluded target and the target of scale changes.This paper proposed a scale adaptive target tracking algorithm for robot based on affine transformation.We defined the corner points,recognized target according to the defined corner points,and recognized the scale changes of target using affine transformation.Compared with relative algorithms,the proposed algorithm can recognize the occluded target effectively,and when the scale of target changes,the proposed algorithm can also recognize the target accurately.The analysis shows that,when there is less than 2 × 104 pixels in an image and less than 25 frames per second in a video stream,the proposed algorithm can be used in real-time target tracking.
出处
《计算机科学》
CSCD
北大核心
2014年第12期280-282,292,共4页
Computer Science
基金
国家自然科学基金课题(91220301)资助
关键词
尺度
机器人
目标跟踪
图像处理
Scale
Robot
Target tracking
Image processing