基于SVM与Meanshift跟踪算法的运动视频目标跟踪  

Moving video object tracking based on SVM and Meanshift tracking algorithm

作　　者：王彬彬 WANG Binbin(Xingzhi College of Xi’an University of Finance and Economics,Xi’an 710038,China)

机构地区：[1]西安财经大学行知学院,陕西西安710038

出　　处：《现代电子技术》2022年第1期56-60,共5页Modern Electronics Technique

摘　　要：文中提出基于SVM与Meanshift跟踪算法的视频运动目标跟踪方法,在体育视频初始图像中选取跟踪目标所处位置,获取跟踪目标周围目标与背景两部分特征向量,使用目标和背景特征向量训练SVM二分分类器,使用分类器分类下一帧视频图像跟踪目标位置与所处背景图像,获取置信图;使用Meanshift跟踪算法在置信图范围内获取当前跟踪目标中心位置,移动目标框和背景框的中心位置到达目标位置,以10%的比例缩放目标框并选择最优者用以适应目标尺寸变化;确定是否已经跟踪到视频最后一帧图像,如果没有跟踪至最后一帧图像,则需使用此时目标像素和背景像素训练新的SVM分类器,跟踪下一帧图像,直至完成整个视频序列图像运动目标跟踪任务。实验结果表明,所提方法可以实时、准确地跟踪视频内运动目标。A video moving object tracking method based on SVM and Meanshift tracking algorithm is proposed. The location of the tracked object is selected in the initial image of sports video to obtain the eigenvectors of both the peripheral objects and the background around the tracked object. The eigenvectors of the object and its background are used to train the SVM binary classifier. The classifier is used to classify the tracked object location and its background in video image of the next frame,so as to obtain the confidence map. The Meanshift tracking algorithm is used to obtain the center position of the current tracked object within the confidence map. The center position of the object box and background box is moved to the object position. The object box is scaled to a proportion of 10% and the optimal one is selected to adapt to the change of the object size,and determine whether the image of the last frame of the video has been tracked. If not,it is necessary to train a new SVM classifier by using the current object pixel and background pixel to track the image of the next frame until the whole moving object tracking task of the video sequence image is completed. The experimental results show that the proposed method can track the moving objects in the video in real time and accurately.

关 键 词：运动目标跟踪 视频图像 SVM分类器 置信图 Meanshift跟踪算法 图像跟踪 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]