一种基于四叉树划分的改进ORB算法

An Improved ORB Algorithm Based on Quad-Tree Partition

原ORB(oriented FAST and rotated BRIEF)算法提取的图像特征点经常出现“扎堆重叠”现象,其分布较为密集且缺乏尺度不变性,因而容易造成图像特征点误匹配的问题。为了解决该问题,提出了一种基于四叉树划分的图像特征点提取算法。首先对图像建立尺度金字塔,然后使用四叉树划分图像并限制划分深度。用加速分段测试的特征(features from accelerated segment test,FAST)算法通过多个检测阈值对划分后的图像进行特征点检测。检测完毕后,根据划分出的子块总数和提取的特征点总数对划分出来的各个子块设置自适应阈值,提取ORB特征点。操作完成后通过采取非极大值抑制的方法筛选最佳特征点,并使用改良后的二元鲁棒独立基本特征(binary robust independent elementary features,BRIEF)算法计算得出特征点的描述符,最后进行特征点匹配。实验结果表明,本文算法提取的图像特征点较原ORB算法提取的效果在均匀程度上得到了明显地提升,冗余重叠的特征点数量减少,且在特征点提取速度方面较原ORB算法的提取速度提高了30%以上。

The image feature points extracted by the original ORB algorithm often appear the phenomenon of?clustering and overlapping?,and their distribution is relatively dense and lack of scale invariance,which easily leads to the problem of mismatching of image feature points.In order to solve this problem,this paper proposes an image feature point extraction algorithm based on quad-tree structure.First,the scale pyramid of the image is built,and then the quad-tree is used to divide the image,and the depth of the partition is limited.The FAST algorithm is employed to detect the feature points of the scaled image by multiple detection thresholds.Second,the ORB feature points will be extracted based on the division of the total sub-block and the total number of the feature points.And then the best feature points are obtained by taking the maximum inhibition method,and the feature points’descriptors are calculated with the help of the improved BRIEF algorithm.Finally the work of feature points matching will be realized.Experimental results show that compared with the original ORB algorithm,the uniformity of feature points extracted by the proposed algorithm in this paper is significantly improved.The number of redundant and overlapping feature points is reduced,and the extraction speed of feature points is improved by more than 30%.