基于方差估计的快速稀疏表示目标跟踪算法研究

针对粒子滤波框架下稀疏表示跟踪算法存在的粒子规模大以及复杂的L1范数最小化计算等问题,提出一种基于方差估计的快速稀疏表示目标跟踪算法。该算法首先在运动估计阶段应用方差估计优化采样粒子的分布;然后利用L2范数代替L1范数对判别目标函数进行稀疏求解计算,并进一步从重构误差的度量形式做出改进来增强L2范数的稀疏性;最后利用在线字典学习(ODL)算法对模板字典进行在线更新。实验结果表明,该算法能够克服干扰实现稳定快速跟踪,提高跟踪效率。

Research on fast sparse representation target tracking algorithm based on variance estimation

A fast sparse representation target tracking algorithm based on variance estimation is proposed for the problems of large particle size and complex L1 norm minimization calculation in the sparse representation tracking algorithm under the particle filter framework. The algorithm firstly uses variance estimation to optimize the distribution of sampled particles in the motion estimation stage, then adopts the L2 norm instead of the L1 norm to sparsely solve the discriminant objective function, and further improves the sparseness of the L2 norm by making improvements from the reconstruction error measurement form. Finally, the online dictionary learning(ODL) algorithm is used to update the template dictionary online. The experimental results show that the algorithm can overcome the interference, achieve stable and fast tracking, and improve the tracking efficiency.