Li Ying, Li Pengcheng, Shen Qiang
Appl Opt. 2014 Oct 1;53(28):6518-26. doi: 10.1364/AO.53.006518.
Tracking a target in infrared (IR) sequences is a challenging task because of low resolution, low signal-to-noise ratios, occlusion, and poor target visibility. For many civil and military applications, the realtime requirement is always a key factor for tracking algorithms to be used. This undoubtedly makes tracking in IR sequences more difficult. This paper presents a real-time IR target tracking under complex conditions based on l1 minimization and compressive features. First, we adopt a sparse measurement matrix to project the high-dimensional Harr-like features to low-dimensional features that are applied to the appearance modeling. This appearance model allows significant reduction in the computational cost of the target-tracking phase. Then, the appearance model is introduced into the framework of the popular l1 tracker. Each IR target candidate is represented by the appearance template based on the structure of sparse representation. Finally, the candidate that has the minimum reconstruction error is selected as the tracking result. The proposed tracking method can combine the real-time advantages of the compressive tracking and the robustness of the l1 tracker. Experimental results on challenging IR image sequences including both aerial targets and ground targets show that the proposed algorithm has better robustness and real-time performance in comparison with two state-of-the-art tracking algorithms.
在红外（IR）序列中跟踪目标是一项具有挑战性的任务，这是由于分辨率低、信噪比低、遮挡以及目标可见性差等原因。对于许多民用和军事应用而言，实时性要求始终是所使用的跟踪算法的关键因素。这无疑使得在红外序列中进行跟踪变得更加困难。本文提出了一种基于l1最小化和压缩特征的复杂条件下的实时红外目标跟踪方法。首先，我们采用稀疏测量矩阵将高维类哈尔特征投影到低维特征，这些低维特征用于外观建模。这种外观模型能够显著降低目标跟踪阶段的计算成本。然后，将外观模型引入到流行的l1跟踪器框架中。每个红外目标候选对象基于稀疏表示的结构由外观模板表示。最后，选择重构误差最小的候选对象作为跟踪结果。所提出的跟踪方法能够结合压缩跟踪的实时优势和l1跟踪器的鲁棒性。在包括空中目标和地面目标的具有挑战性的红外图像序列上的实验结果表明，与两种最先进的跟踪算法相比，所提出的算法具有更好的鲁棒性和实时性能。
