摘要
在随机树算法的基础上,提出一种内点回收算法。该算法修改了特征点匹配机制,不再只将当前帧特征点与后验概率最大的关键帧特征点进行匹配,通过构造可能匹配点集合,利用特征纹理在图像上的空间信息从集合中选择最佳匹配点,有效利用当前帧提取到的所有特征点,提升重复纹理图像的匹配效果。
There are lots of outliers when traditional feature matching methods are used to match two images which contain characters,symmetrical objects or other objects contained repetitive texture.The number of outliers is so large in this case that the number of inliers reduces sharply after RANSAC（Random Sample Consensus）,which affects the accuracy of camera pose.An inliers retrieval method based on randomized tree is presented.For each point on the current frame,the proposed method creates a set which contains all points on key frame that can match with this point,and utilizes the spatial information of points on key frame to choose the best matching point.Experimental results show that the proposed retrieval method can improve the matching effect when the processed images contain repetitive texture.
基金
国家高技术研究发展计划(863计划)资助项目(2008AA12A220
2009AA012106)
国家自然科学基金资助项目(60827003
61072096)
关键词
特征识别
增强现实
重复纹理
feature matching
augmented reality
repetitive texture
