基于边缘特征的自动运动对象分割  

Automatic Moving Object Segmentation Based on Edge Features

作　　者：张晓燕[1] 赵荣椿[1] 马志强[2] 钱渊[2] 

机构地区：[1]西北工业大学计算机学院,西安710072 [2]空军工程大学电讯工程学院,西安710077

出　　处：《电子与信息学报》2007年第2期460-464,共5页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(60141002);南昌航院测控中心开放实验室基金(KG200104001)资助课题

摘　　要：为了准确分割出视频场景中的运动对象,该文提出了一种基于边缘特征的运动对象分割及跟踪算法。首先对相邻帧进行自适应变化检测,得到相邻帧二值差分图像。结合当前帧Canny算子检测的边缘图像,获得运动对象的初始边缘模板。其次对运动对象的运动分为快变和慢变两部分进行跟踪并更新运动对象的边缘模板。最后对运动对象的边缘模板进行数学形态学处理得到运动对象的外轮廓,使用梯度向量流场作为外力的改进活动轮廓算法收缩获得运动对象准确的闭合轮廓曲线。该算法对运动对象的整体运动和局部形变都有很强的鲁棒性,能够得到运动对象准确的轮廓,并且对复杂背景有很好的适应性。In order to segment the moving objects accurately, an edge based moving objects segmentation algorithm is proposed in this paper. First of all, the initial edge model of thc moving object is obtained on the basis of the adaptive change detection method and Canny edge detection. Secondly, the binary edge model is tracked and updated to accommodate rotation and changes in shape of the tracked object. Finally, the outer contour of the moving object is determined based on the mathematical morphological operation and an improved active contour which uses the gradient vector as the external force guides the initial outer contour moving to the actual contour. The proposed algorithm is robust to the entire motion and local deformation of object, adaptable to the complex background and can obtain the accurate moving object contour.

关 键 词：运动对象 变化检测 边缘检测 数学形态学 活动轮廓 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]