结合运动目标轮廓提取的均值漂移算法  

Mean-shift tracking algorithm combined with target contour extraction

作　　者：韩兆福[1] 曾令伟[1] 黄晨[1] 戴贤妙[1] 

机构地区：[1]装甲兵工程学院控制工程系,北京100072

出　　处：《激光与红外》2010年第11期1258-1262,共5页Laser & Infrared

摘　　要：提出一种结合运动目标轮廓提取的改进均值漂移算法。对相邻4帧的图像两两配准之后进行差分相乘运算,再进行形态学处理获得理想的目标区域,经权值计算之后在该区域内建立目标颜色直方图模型,计算均值漂移向量进行目标跟踪,当相似性系数小于某个阈值时更新目标模板模型,以适应目标的形变等情况。实验结果表明,改进后的算法具有较好的抗背景干扰能力,能够有效和准确地跟踪目标,而且对跟踪不规则和形变目标有很好的适应性,提高了均值漂移算法的跟踪性能。This paper aims to propose an advanced mean-shift algorithm which based on contour extraction of moving target.Four adjacent frames of images are obtained,the first frame minus the third frame and the second frame minus the fourth frame after registration,and then multiplied by their differences.Morphological processing method was taken to obtain an ideal target area where target color histogram is established after weight calculation.After that,this algorithm calculates the mean-shift vector to track the target.Target color histogram is updated in order to adapt the target deformations when Bhattacharyya coefficient is lower than threshold.The experimental results show that the algorithm with good capability of resisting background interference,and also tracking targets effectively and accurately.This algorithm has strong robustness to both structure irregularity and deformation,it enhance the tracking capability of mean-shift algorithm.

关 键 词：均值漂移 轮廓提取 差分相乘 目标颜色直方图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]