提出了一种包含区域信息的Snake模型用于运动目标检测与跟踪.在通常情况下,基于区域信息的跟踪方法对背景光线的微小变化、位置的微小移动较为敏感,而基于边缘信息的跟踪方法则难以对边缘模糊的图像取得满意的跟踪效果.在算法中同时引入这两种信息,边缘信息使得算法快速而鲁棒性好,区域信息可以对边缘模糊的图像取得正确的跟踪效果.使用双差分图像设计了自动初始化的方法来实现视频的自动跟踪.同时,对目标的下一步运动位置增加了一个预测环节来加快主动轮廓模型的收敛速度.该算法的每帧计算时间一般小于0.1s,能应用于实时系统.%A novel Snake model with region information is proposed to detect and track moving objects. Generally, the region-information-based approach is sensitive to illumination changes and small movement in the background, while the edge-information-based approach often obtains incorrect results for ambiguous images. The two types of information are introduced in computing the image force. Edge-information-based features make the algorithm fast and robust, and region information makes the active contour energy function obtains correct results for ambiguous images. Furthermore, an automatic contour initialization method using double difference images is given to meet the requirement of video sequence tracking. Meanwhile, a simple forecast section is added to estimate the position of the contour in the algorithm so that it can improve the convergence speed of the active contour. Experimental results show that the computation time of the algorithm is less than 0.1 s/frame. And it can be applied to a real-time system.
展开▼
机译：提出了一种包含区域信息的Snake模型用于运动目标检测与跟踪.在通常情况下,基于区域信息的跟踪方法对背景光线的微小变化、位置的微小移动较为敏感,而基于边缘信息的跟踪方法则难以对边缘模糊的图像取得满意的跟踪效果.在算法中同时引入这两种信息,边缘信息使得算法快速而鲁棒性好,区域信息可以对边缘模糊的图像取得正确的跟踪效果.使用双差分图像设计了自动初始化的方法来实现视频的自动跟踪.同时,对目标的下一步运动位置增加了一个预测环节来加快主动轮廓模型的收敛速度.该算法的每帧计算时间一般小于0.1s,能应用于实时系统.%A novel Snake model with region information is proposed to detect and track moving objects. Generally, the region-information-based approach is sensitive to illumination changes and small movement in the background, while the edge-information-based approach often obtains incorrect results for ambiguous images. The two types of information are introduced in computing the image force. Edge-information-based features make the algorithm fast and robust, and region information makes the active contour energy function obtains correct results for ambiguous images. Furthermore, an automatic contour initialization method using double difference images is given to meet the requirement of video sequence tracking. Meanwhile, a simple forecast section is added to estimate the position of the contour in the algorithm so that it can improve the convergence speed of the active contour. Experimental results show that the computation time of the algorithm is less than 0.1 s/frame. And it can be applied to a real-time system.
展开▼
