摘要
针对压缩跟踪算法分类器更新比较盲目导致跟踪准确性下降的问题,提出一种基于最优特征更新分类器的压缩跟踪算法。在原始算法基础上引入确定性测量矩阵,提高压缩感知性能;为了避免被污染样本影响分类器参数更新,不使用所有压缩特征更新分类器,而是在线筛选出最优的压缩特征更新分类器。同时,利用相邻两帧目标仿射变换使跟踪窗口可随目标变化实时更新,实现多尺度跟踪。实验结果表明,算法可有效抵抗光线、遮挡、尺度等因素对跟踪的影响,具有更高的稳定性和更好的鲁棒性,且满足实时性要求。
Aiming at the problem that the classification accuracy of the compression tracking algorithm classifier was relatively blind,the compression tracking algorithm based on the optimal feature updating classifier was proposed. On the basis of the original algorithm, the deterministic measurement matrix was introduced to improve the compression perceptual performance. In order to avoid the contaminant samples affect the updating of the classifier parameters and not use all the compression features to update the classifier,the optimal compression feature updating classifier was screened out. At the same time,the use of adjacent two-frame target affined transformation so that the tracking window could be real-time changes with the target changes to achieve multi-scale tracking. The experimental results showed that the proposed algorithm could effectively resist the influence of light,occlusion and scale on the tracking,and had higher stability and better robustness,and meet the real-time requirements.
出处
《计算机应用与软件》
北大核心
2018年第2期206-211,共6页
Computer Applications and Software
关键词
压缩感知
确定性测量
矩阵特征
置信度
巴氏系数
仿射变换
Compression perception
Deterministic measurement matrix
Feature confidence
Bhattacharyya coefficient
Affine transformation