基于Snake算法的深度图像人体目标跟踪  

Human object tracking in depth image based on Snake algorithm

作　　者：杨晗芳 张国山[1] 王欣博[1,2] 凌朝清 李妍 

机构地区：[1]天津大学电气与自动化工程学院,天津300072 [2]天津市三特电子有限公司,天津300384

出　　处：《天津理工大学学报》2014年第5期41-45,共5页Journal of Tianjin University of Technology

基　　金：国家自然科学基金(61074088)

摘　　要：基于彩色图像的人体跟踪算法鲁棒性不高的主要原因是对目标进行跟踪时,受到光照变化、复杂背景、物体遮挡等因素的影响.针对此问题本文利用Kinect采集深度图像进行人体目标跟踪.首先在深度图像中通过用户索引检测出人体目标,可方便地去除图像中复杂背景的干扰.然后利用基于角点的自动初始化方法得到人体的轮廓信息,再结合Snake算法实现人体目标跟踪.最后将该算法与基于深度图像的Camshift算法进行对比分析.结果表明,在室内应用Snake算法不受灯光和复杂背景等因素的影响,能对人体目标进行实时跟踪,且比Camshift算法具有更强的抗干扰能力,跟踪更准确.The main reason why the robustness of human object tracking algorithm based on color image is not high is that when tracking objects,affected by illumination change,shadow,object occlusion and so on.To solve the problem,we use depth image collected from Kinect to achieve human object tracking.Firstly,user index can be used to detect human object in depth image,and interference of complex background can be easily removed.Secondly,we extract human contour information through automatic initialization method based on corner,then Snake algorithm is combined to achieve human object tracking.Finally,we compare and analyze the algorithm with Camshift algorithm based on depth image.Results show that Snake algorithm is not affected by light change and complex background and other factors in indoor,and it can achieve human object real-time tracking,Snake algorithm is more accurate with stronger anti-interference ability than Camshift algorithm.

关 键 词：KINECT 深度图像 SNAKE 连续自适应均值漂移(Camshift) 人体目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]