基于奇异值分解的图像目标跟踪算法

传统相关跟踪方法是利用模板图像与目标图像对应像素的灰度差异信息进行跟踪，它对旋转变化敏感，且存在跟踪累积误差，容易导致模板漂移而丢失目标。文中提出基于奇异值分解的跟踪算法，算法首先建立模板图像训练集合，利用奇异值分解方法，张成模板图像特征空间，然后求出模板图像在特征空间里的投影值，代替传统算法中灰度对两幅待匹配图像进行的全局搜索定位。在进行投影值间的相似性度量时，欧氏距离同等对待所有的特征向量不移合理，文中采用了一种鲁棒估计方法，可以对不同距离的值做不同处理。匹配跟踪实验效果良好。

Image matching method based on singular value decomposition

In this paper, a new approach for tracking rigid objects is presented. It builds on and extends on SVD method and Robust estimation techniques. Because of the presence of the difference of the gray value and the distortion between the object image and the reference image, traditional matching method will be degraded. We need more useful characters. Eigenspace approach is a promising candidate for an appearance-based object representation. In this paper, the tested object images are decomposed first, the eigen spaces of the images are obtained, then they are applied as the characters of the object images in the matching process. Another point in this paper is the distance measurement. We think that the least-squares method has a number of problems, so a robust estimation is adopted to make the similarity measurement. These techniques are used to track objects over long image sequences, the experimental results show that this image matching method could bear affine image motion and image distortion to some extent and it is very promising.