基于增量学习的稀疏表示仿射目标跟踪算法研究

Research on Affine Object Tracking Using Sparse Representation Method Based on Incremental Learning

针对传统目标跟踪算法在背景复杂.目标形态和光照条件剧烈变化情况下跟踪效果不佳的问题,提出了一种新的目标跟踪算法.该算法在粒子滤波框架下用仿射变换和Gabor特征表示图像,用模板字典稀疏表示候选目标,并用增量学习算法对模板字典进行更新.试验部分将该算法与其他跟踪算法在Matlab平台上进行比较,试验结果表明该算法具有鲁棒性强、跟踪效果好的优点.

For improving the poor effect of traditional tracking algorithms under the condition of complicated background,significantly varying appearance and illumination,this paper proposes a novel algorithm for visual tracking.In the particle filter framework,images are represented by the use of the affine transformation and the Gabor feature. The candidate target is sparsely represented by a template dictionary and the template dictionary is renewed through the incremental learning algorithm. In the experimental section,our approach is compared with other visual tracking approaches in Matlab platform. Experimental results show that our approach has better performances and is more robust than other tracking approaches.