Abstract: To solve the problems of poor robustness and low effectiveness of visual tracking in complex scenes, this paper proposes a target tracking algorithm based on adaptive observation weight in Bayesian framework. Firstly, a weighted observation model is established via linear visual tracking representation. Then an iterative optimization algorithm is put forward to adaptively update the weight matrix to eliminate negative influences of observation outliers. Finally, effective likelihood evaluation function is adopted to capture the target accurately. The experimental results show that the proposed algorithm outperforms other state-of-the-art tracking algorithms in tracking accuracy and robustness.

Key words: visual tracking, linear representation, on-line updating, outliers, adaptive weight matrix

摘要： 针对视觉跟踪在复杂场景中跟踪精度较低和鲁棒性较差的问题，在贝叶斯框架下提出了一种自适应观测权重的目标跟踪算法。通过视觉跟踪中的线性表示模型构建出一种加权观测模型；提出一种基于迭代加权的模型优化算法，利用在线更新的自适应权重矩阵消除观测离群值对跟踪有效性的影响；最后，采用有效的似然评估函数实现对目标准确、鲁棒的跟踪。实验结果表明，该算法在跟踪精度和鲁棒性方面都优于现有的一些跟踪算法。

关键词: 视觉跟踪, 线性表示, 在线更新, 离群值, 自适应权重矩阵