基于压缩感知的在线多示例学习目标追踪

近年来提出的多示例学习算法在一定程度上能够克服模板漂移问题。然而,在线学习需要获取足够多的有用数据才能达到稳定的追踪效果,但是这却增加了算法的复杂度。为了解决这一问题,在压缩感知理论的基础上,运用随机观测的方法对多尺度图像特征进行降维,提取的这些低维特征中包含大量的有用信息。因此,我们提出的算法是先利用压缩感知理论提取目标特征之后,再使用在线多示例学习算法分类器对这些特征进行分类从而实现目标的稳定跟踪。通过对不同的图像序列进行实验,结果表明基于压缩感知的在线多示例学习算法对实时的目标追踪有很好的适应性。

Visual Tracking with Multiple Instance Learning Based on Compressive Sensing

Visual tracking is one of the most popular research topics in the domain of computer vision. It is a challenging task to develop an effective and efficient tracking algorithm because of template drift problems. To alleviate the drift, the multiple instance learning (MIL) method has been applied to target tracking. However, there must be a sufficient amount of useful data for online MIL to learn at the outset, which actually increases the computational complexity. In this paper, an effective tracking algorithm is proposed which uses an online MIL based on the compressed appearance model to accomplish object tracking. In order to decrease the computational complexity and obtain sufficient data for online learning adaptive appearance model, Features are extracted by non-adaptive random projections of the multi-scale image feature space based on compressive sensing theories. The experimental results on various videos show that the proposed method has a satisfactory performance in real-time object tracking.