基于ML和L2范数的视频目标跟踪算法

摘要

目标跟踪是计算机视觉领域的一个具有挑战性的问题，本文提出了一种基于ML （最大似然）估计和L2范数的视频目标跟踪算法。建立基于稀疏限制的ML模型，给样本中的异常像素分配较小的权值，减少异常像素对跟踪算法的影响。利用L2范数最小化进行稀疏编码求解。采用贝叶斯估计得出目标跟踪结果。与其他典型算法相比，本算法降低了计算的复杂度，对遮挡，旋转，尺度变化，光照变化等异常变化具有较强的鲁棒性。%The tracking of target is a challenging issue in computer vision .In this paper ,we propose a visual object tracking algorithm based on ML estimation and L2-norm .Firstly ,the model of sparsity constrained ML is established .Abnormal pixels in the samples will be assigned with low weights to reduce their affects on the tracking algorithm .Then ,L2-norm minimization is used to solve the sparse coding .Finally ,the object tracking results is obtained using Bayesian MAP estimation .Compared with other popular methods ,our proposed method reduces the computational complexity and has stronger robustness to abnormal changes (e .g .occlu-sion ,rotation ,scale change ,illumination ,etc .)