基于ORB和角点方向夹角约束的快速图像配准方法

High Speed Image Registration Algorithm Based on ORB and Corner Angle Constraint

针对图像间存在不同形变、光照等情况下配准难度大的问题,基于尺度不变特征变换(SIFT)的配准方法计算量大、无法满足实时性的要求,提出一种基于Oriented FAST and Rotated Brief(ORB)与角点方向夹角约束的快速图像配准方法。首先在两幅图像中分块提取ORB特征点,采用一种基于双阈值的汉明距离进行特征点匹配,针对随机抽样一致性(RANSAC)算法无法剔除错误匹配的特征点,以角点方向夹角一致性为约束条件,有效剔除误配点;然后再用RANSAC算法计算出最佳变换矩阵,完成图像配准。实验表明,采用具有不同形变、光照等情况下的3组图像,该方法不仅能很好地配准图像,并且3组图像配准的平均时间为80.399 ms,不到SIFT配准方法所需时间的1/20,兼顾了图像配准的有效性和实时性。

For the image regstration is difficult under the condition of the different deformation, illumination and others, the registration algorithm based on scale invariam features trandonn （SIFF）is complex, and it dees＇t meet the real-time requirement. So, a high speed registration algorithm is proposed based on Oriented FAST and Rotated Brief（ORB） and comer angle constraint. Firstly, ORB feature points are extracted in two partitioned images, then the feature points are matched with double threshold＇s hamming distance, for random sample consensus （RANSAC） can＇t reject the mismatched feature points. Comer angle constraint is proposed to reject the mismatched feature points, and then calculate the optimal transform matrix with RANSAC algorithm to finish the image registration. Experimental resuits show that, with three gronps of images with different deformation and others,the algorithm can rectify the image, and its average time to rectify the three groups of images is 80.399 ms, less than 1/20 of the SIFTs average time, it matches both the validity and time.