提出了一个基于形变多样相似性的空间正则化相关滤波跟踪算法。在核相关滤波(KCF)跟踪算法基础上引入了空间正则化权重和子网格检测方法,利用形变多样相似性匹配算法构建了目标重检测模块,利用主成分分析(PCA)算法和k维树一致近似最近邻(TreeCANN)算法解决了匹配算法中的最近邻搜索问题;通过自适应模板更新策略,解决了遮挡情况下模板误更新问题。实验结果表明,所提算法的精确度得分为0.825,成功率得分为0.625,相比KCF算法分别提升了18.5%和31.0%。所提算法能较好地解决目标尺度变化、遮挡、快速运动、旋转和背景杂乱情况下的跟踪问题,具有广泛的应用前景。

The spatial regularization correlation filtering tracking algorithm based on deformable diversity similarity is proposed. The spatial regularization weight and the sub-grid detection method are introduced on the basis of the kernelized correlation filtering (KCF) tracking algorithm. The target re-detection module is constructed with deformable diversity similarity matching algorithm, the nearest neighbor search problem in the matching algorithm is solved via the principal component analysis (PCA) algorithm and the k-direction tree coherence approximate nearest neighbor (TreeCANN) algorithm. Through the adaptive template updating strategy, the problem of false updating of the template under occlusion is solved. Experimental results show that the precision score and success score of the proposed algorithm are 0.825 and 0.625, which are 18.5% and 31.0% higher than those of the KCF algorithm, respectively. The proposed algorithm can better solve the tracking problems of the target scale variation, occlusion, fast motion, rotation and background clutter, showing a wide range of application prospect.