基于压缩感知的实时在线目标追踪

在正负样本区域随机抽取了不同尺度下图像的局部二值模式(LBP)特征,将高维的特征信息投射到低秩的压缩域,并据此建立了表观模型.使用一个随机稀疏测量矩阵来压缩前景和背景目标.将追踪问题转化成为了一个使用朴素贝叶斯分类器的二元分类问题.所提方法可以较快速、实时地在线追踪目标,同时解决了目标尺度变化、遮挡问题.

Real-Time online target tracking based on compressed sensing

We randomly extracted the feature of local binary patterns(LBP) with different sizes within both positive samples and negative samples.The high-dimensional feature information was projected onto the low rank compression domain based on which the characterization model was established.Then,it compressed samples of foreground and the background targets by using the same random sparse measurement matrix.Finally,the tracking task was formulated as a binary classification via a Naive Bayes classifier.The experiment showed that the proposed method could track the target quickly and constantly.Furthermore,it could also solve the problem of multi-scale change and occlusion issue at the same time.