改进型粒子滤波与稀疏表达的自适应图像跟踪算法

AN ADAPTIVE VISUAL TRACKING ALGORITHM FOR IMPROVED PARTICLE FILTRATION AND SPARSE REPRESENTATION

针对各种复杂场景下视频序列目标跟踪算法对目标外形、部分遮挡、灰度等参数敏感的问题,提出基于改进型粒子滤波与稀疏表达的自适应图像跟踪算法。通过建立过完备基,建立样本集合以及提取特征集合得到过完备集合。使用正交匹配跟踪算法求取稀疏系数,提出稀疏度计算公式用以计算各区域的匹配值,使用各匹配值求取目标位置。实验结果显示,算法能够稳健、高效地跟踪运动目标,相对各种运动跟踪算法,算法运行速度快,鲁棒性高,能够完成多种复杂环境下的跟踪任务。

To solve the problem that in various complicated scenes video sequence object tracking algorithm are sensitive to such parameters as object figure, partial coverage, grey grade and so on, the paper proposes an adaptive visual tracking algorithm based on improved particle filtration and sparse representation. By building over-completed bases, it builds sample sets and extracts feature sets to obtain over-completed bases. It utilizes orthogonal matching pursuit to get sparse coefficients, provides a sparse degree computational formula to calculate matching values of various regions and use them to solve the object＇ s position. Experiment results show that the algorithm can steadily and efficiently track moving objects. Compared with various other moving tracking algorithms, the proposed one runs faster, is more robust and can complete the tracking tasks in a number of complicated environments.