为快速稳定地匹配视频序列，并考虑SVD算法的高效性，根据视频序列的特点，对SVD匹配算法进行改进，提出了一种适合视频序列的匹配算法。该算法使用Harris角点检测算子检测兴趣点，使用有向模板提取具有旋转不变性的特征，并通过引入颜色加权法改进SVD算法中的相似性度量函数。同时，又提出一种基于运动一致性约束的误配点剔除方法，首先拟合匹配点间的运动模型，然后自适应地调整参数将错误的匹配点剔除。该算法使用有向模板消除图像间旋转变换的影响，使用颜色特征降低兴趣点匹配时的不确定性，通过运动一致性约束降低误配点数量。实验结果表明，该算法在图像间存在旋转变换关系和不同的光照条件时都可以获得很好的匹配结果，特别是在图像间基线距离较大时仍能得到大量的匹配点并具有很高的正确匹配率，能很好满足实际需要。

In order to match video sequence images efficiently and considering the effectiveness and efficiency of the SVD matching algorithm, this paper describes a more advanced SVD multi view video image matching algorithm based on a new type of invariant feature and outlier rejection strategy. The features are located at Harris corners and oriented using a blurred local gradient. This defines a rotationally invariant frame which is sampled as a feature descriptor. The color information is introduced into the distance function of SVD matching algorithm. The outliers are rejected based on motion consistency. For motion consistency a motion model was fit first between the matched points, and the threshold was set adaptively using the deviations of the motion to reject the outliers. This algorithm uses oriented patch to adapt to the effect of rotation, color features to remove the uncertainty, and motion consistency to increase the correct match ratio. The experimental results show that the algorithm can obtain lots of matching points between wide base line images, even if there is rotational transformation and different lighting conditions between two images, which proof that the algorithm has practical value.