仿射变换在压缩感知跟踪中的应用

针对跟踪中目标尺度变化和旋转问题,将仿射变换和应用到压缩感知跟踪中.首先,以上一帧的跟踪结果为均值,以一定的标准差按照高斯分布,随机生成不同尺度和旋转角度的候选框;然后,通过仿射变换将其转换至直角坐标系中,通过多尺度滤波得到目标在不同尺度下的高维特征向量,采用压缩矩阵将高维特征向量降维至低维空间;最后,将低维特征向量通过贝叶斯分类器选取具有最大响应的候选位置作为目标的跟踪位置.在此基础上分别提取正负样本来更新分类器参数,从而实现持续稳定的跟踪.实验结果表明,该算法能够较好地解决压缩感知跟踪中的目标旋转和尺度变化问题.

Application of affine transform in compressed sensing tracking

The affine transform is employed to solve the scale change and rotation in in compressed sensing tracking. Firstly according to the Gauss distribution of a certain variance, a number of candidates are produced based on the former tracking result. Then the patch is transformed into the orthogonal coordinate via affine transformation. And the high-dimension feature vector is acquired through the multiple-scale filter. The high-dimension feature vector is compressed through the compressive matrix. Finally, the low-dimension feature vector is passed through the bayes classifier and the candidate with the highest response is recognized as the tracking result, on the basis of which the positive and negative samples are extracted to update the parameters of the naive bayes classifier. Experimental results show that the proposed algorithm can well cope with the scale variation and rotation in compressed visual tracking.