一种采用排名机制的分块在线压缩跟踪算法

An Online Compressed Block Tracking Algorithm Using Ranking Mechanism

基于分类学习的目标跟踪在面对环境中光照变化、目标姿态变化以及遮挡等复杂环境下容易出现漂移问题,为此提出一种基于分类器融合的压缩感知目标跟踪算法。使用压缩感知理论分块提取目标压缩特征,根据贝叶斯后验概率对特征进行筛选以构建目标模型,并提出一种二阶段样本搜索方法,通过粗搜索缩小样本的搜索范围,利用基于分类器排名的细搜索方法精确地找到目标的位置。实验表明,该算法与当前主要的算法相比具有较高的跟踪精度,以及良好的鲁棒性和实时性。

The tracking methods based on classification usually result in drifting in complex environment owing to the illumination variation in environment,object posture changing and occlusion. A compressive sensing moving object tracking technique was presented based on classifier fusion. Firstly,compressed features of the object were extracted through the theory of compressive sensing,and then selected the features to construct the object model according to the Bayesian posterior probability. Finally,a two stage searching method was proposed,which narrowing the search range of samples through rough search,then finding the exact position of the object through fine search based on the ranking of classifier. Experiments and comparisons with other methods which are popular or classical,demonstrate the improvements in tracking accuracy,robustness and real-time performance.