一种改进的基于SURF的视频帧间匹配方法

An Improved Matching Method in Video Frames Based on SURF

针对视频连续帧间匹配不准确、错误率高、匹配速度慢的问题,提出了一种改进的基于SURF(Speeded Up Robust Feature)特征点的匹配方法。按照SURF算法进行特征点检测和描述;对视频连续帧利用改进的最近邻与次近邻的比的方法进行双向匹配,在匹配时仅在以相应位置为中心的邻域内寻找最近邻点和次近邻点,根据最近距离与次近距离的比值与预先设定阈值的比较结果确定是否接受这一匹配点对;用RANSAC(Random Sample Consensus)方法建立变换矩阵模型剔除错误匹配点,得到精确匹配的特征点对,完成匹配过程。在经典的视频数据集上进行实验,实验结果表明该方法不仅提高了视频连续帧间匹配的正确率,同时使匹配时间相对缩短了一半左右,显著提高了匹配效率,证明了算法的有效性。

With the problem of inaccurate matching,high error rate and lowspeed in video frames,an improved matching method based on SURF（ Speeded Up Robust Feature） is presented.The SURF features are detected and described. The improved method of ratio between the nearest and the next nearest neighbor is used for bidirectional matching.When matching,the nearest and next nearest neighbor points are searched only in the neighborhood of the corresponding points and the two matching points are accepted according to the comparison results between the distance ratio and the present threshold. The RANSAC method is applied to build the transformation matrix model to removing the error matches and get the exact match,completing the match process.The experiment is carried out on the classic video dataset,and the result shows that the method can improve the matching accuracy,and the matching time is relatively shortened by about half,significantly improving the matching efficiency and verifying the effectiveness of the algorithm.