Abstract: By analyzing the distribution of the particles sampled from particle filter and their difference with the true states of the moving target, a robust sparse tracking method based on online logistic discriminate learning is proposed. The discrimination of the moving target and backgrounds is enhanced by the discriminating and updating procedures of the logistic function. The proposed method is more efficient than the original sparse tracking since the particles with little relevancy are rejected in advance of the L1 optimization. Experimental results demonstrate that, with 5 state-of-the-art tracking under 4 challenging video sequences, the proposed tracking method obtains the most robust results, and its running time is almost equivalent with the BPR-L1 tracking method.

Key words: visual tracking, sparse tracking, online discriminate learning

摘要： 通过分析经典稀疏视觉跟踪算法在粒子滤波框架下的采样粒子分布与运动目标真实状态的差异，提出了一个基于在线判别分析的改进稀疏视觉跟踪算法。该跟踪算法通过在线逻辑斯蒂判别分析模型及其更新过程，自主获取运动目标的实时状态与变化，增强运动目标与背景信息之间的可判别性。同时，实现对采样粒子的预先筛选，尽量排除与运动目标差异大的粒子，以提高跟踪算法的鲁棒性，同时减少L1优化求解的次数从而提高算法的执行效率。与5个高水平跟踪算法在4段公开视频上的实验结果表明，提出的算法能够长时间鲁棒地对运动目标进行跟踪，同时相对典型稀疏跟踪算法而言，明显地降低了计算复杂度。

关键词: 视觉跟踪, 稀疏跟踪算法, 在线判别分析