摘要
在研究基于支持向量机(Support Vector Machine,SVM)和均值漂移(Mean shift)跟踪算法基础上,为了提高目标跟踪的精确度和效率,提出了一种改进的目标跟踪算法。首先将目标跟踪转化为目标和背景的二分类问题,提取目标和背景的特征信息,采用粗糙集理论中的属性约简算法提取它们的有效特征,将目标和背景的特征分别作为正负样本训练SVM,得到背景与目标的总体分类器。借助总体分类器来区分临近下一帧中的目标和背景,并得到置信图;最后通过Mean shift算法找到置信图的峰值,从而获得目标的新位置。实验结果表明,改进方法能有效地提高目标的跟踪精度,同时加快了跟踪速度。
In order to improve the robustness of the tracker, an improved algorithm based on SVM and Mean shift was proposed. The improved method was used to train one SVM classifier for per frame, then extract the characteristics of the target and background information. Using attribute reduction algorithm of rough set theory to extract effective features, we trained the characteristics of the target and background as positive and negative samples for SVM classifier. The ensemble classifier was used to distinguish the target from the background in the next frame and pro- duce a confidence map. The peak of the map was obtained as the new position of the object by mean shift algorithm. The experimental results show that this method can effectively improve tracking accuracy, and it can speed up the tracking speed.
出处
《计算机仿真》
CSCD
北大核心
2014年第6期347-351,共5页
Computer Simulation
关键词
支持向量机
均值漂移
目标跟踪
粗糙集理论
置信图
Support vector machine ( SVM )
Mean shift algorithm
Object tracking
Rough set theory ( RS )
Confidence map