基于稀疏贝叶斯学习的相关向量跟踪

The Sparse Bayesian Learning Based Revelance Vector Tracking

结合稀疏贝叶斯学习方法和支持向量跟踪(SVT)原理,提出了相关向量跟踪(RVT)。由于跟踪系统事先学习到了目标的“知识”,故匹配发生在候选图像块与先验知识之间,而不必考虑模板更新。相关向量有比支持向量更稀疏的性能,所以相关向量跟踪比支持向量跟踪有更快的帧处理速度。另外,为了解决由于运动导致目标尺寸发生变化的匹配跟踪问题,采用了灰度真方图特征,引入了运动预测和变尺寸采样的方法。上述性能和方法在实验中得到了证实。

Combining sparse Bayesian learning with the principle of the support vector tracking(SVT),the relevance vector tracking (RVT) is presented.Because of the matching between the candidated image patch and the prior knowledge,it need not the template updating.And the process time for each frame of the RVT's is faster than the SVT's by reason of the more sparse property of the former.In addition,the gray histogram character is adopted and the motion prediction and the size-alterable sampling methods are used to solve such a matching tracking problem that the size of the object is updated consecutively due to the motion.The above property and the methods have been confirmed in the experiment.