基于Haar-like特征的实时L1-跟踪算法

稀疏表示技术已成功应用于视觉跟踪,但是仍然存在跟踪算法效率低的问题。提出一种基于Haar-like特征的视频跟踪算法,该算法是基于粒子滤波框架的L1-跟踪算法,其特点是运用Haar-like特征及特征块的思想对完备基进行重新构造。将正负小模板由单个像素改为像素块,降低稀疏表示中过完备基的维数,大幅减少稀疏矩阵的计算量;同时,在保证跟踪质量的前提下适当减少目标模板数量,减少稀疏计算的次数,并控制模板更新频率。实验结果表明,所提算法能大幅提高跟踪的实时性,同时很好地解决了跟踪问题中的短时间遮挡、目标物体的形变以及光照变化等问题。

Real-time L1-tracker Based on Haar-like Features

In order to solve the problem of real-time in L1-tracker algorithm under the framework of particle filter,this paper proposed an improved L1-tracker algorithm based on Haar-like features.Firstly,the complete dictionary is reconstructed by the Haar-like features and feature blocks.The single pixels are replaced by the pixel blocks to make up positive and negative trivial templates.Then,the dimensions of over-complete dictionary are reduced by sparse representation and the calculation amount of the sparse matrices is significantly reduced.Secondly,the number of the target template is reduced in order to decrease the calculation of sparse representation.Finally,the updating frequency of the templates is controlled by experiential value.Experimental results demonstrate that the proposed algorithm can significantly improve real-time of L1-tracker algorithm and is effective under short time occlusion,the deformation of target and illumination changes.