混合线性模型在模板跟踪中的应用

模板跟踪在计算机视觉上已经有了很广泛的应用。利用图像的单应性特征，构建目标的运动模型，找出观察数据与目标运动参数的关系，估计目标的运动参数，实现目标跟踪的目的。本文提出了一种基于混合线性模型的模板跟踪方法，提取目标的运动参数和表观特征，建立数据集，利用全监督学习的方法计算出两者的映射关系，从而实现对目标进行有效跟踪。这种方法既克服了由单一线性模型造成的非线性误差，又减少了由于计算非线性模型的计算量，提高了目标跟踪的精度。此外，本文还提出了一种快速学习的计算方法，克服了混合模型中每个子空间由于学习样本少而容易受到噪声影响的缺点，不仅增加了系统的鲁棒性，而且减少了计算量。实验结果表明该算法具有良好地跟踪效果。

Title Mixture Hyperplanes Approximation for Template Tracking

Template tracking has been extensively studied in Computer Vision with a wide range of applications. The motion model is constructed by homography, and the object is tracked by approximating the motion model based on the relationship of the observed data and the motion parameters. This paper proposes a method based on finite mixtures of generalized linear regression models to perform robust, real-time tracking from a stationary camera. A fast learning strategy is discussed as well, which will improve robustness against noise. Moreover, the performance and the stability of Mixture Hyperplanes Approximation are demonstrated and evaluated on a set of challenging image sequences.