为了提高多目标图像的ORB匹配的正确率, 提出一种基于Mask R-CNN的图像ORB去除误匹配方法, 该算法首先通过Faster R-CNN方法对图像进行识别, 运用区域推荐网络得到矩形框标注的感兴趣区域和类别标签, 该步骤可以得到感兴趣区域的预测类别和坐标信息, 并且通过全卷积网络卷积层进行像素级别校正, 得到像素级别的目标所属类别, 然后进行目标分割。最后在原有ORB特征点匹配基础上, 剔除两幅图像中相同目标分割区域以外的误匹配点。为了验证该方法的有效性, 对传统ORB匹配与基于本文方法的ORB匹配进行了仿真实验。改进后的算法, 使得在多目标环境下的目标的匹配精度提高了约18.6%, 结果表明, 本文算法较传统的ORB匹配算法的精度有一定提高。

This paper presents an elimination image ORB mis-matching method based on Mask R-CNN in order to improve the accuracy of multi-target image matching ORB. First of all, the algorithm identifies the image by faster R-CNN method, and it uses Region Proposal Network to obtain the interest area and category labels which are marked by rectangular box. We can get the forecast category and coordinate information of the interest area by this step. And then we correct the pixel level by the convolution layer of the Fully Convolutional Networks, and get the category of target belonging to the pixel level and segment target. Finally, the mis-matching points outside the same target segmentation area are eliminated in the two images on the basis of the original ORB matching. At the same time, some simulation experiments which use the traditional ORB matching method and the ORB matching based on this paper is carried out to verify the effectiveness of this method. The results show that the algorithm is much more accurate than that of the traditional ORB matching algorithm. The improved algorithm makes the matching accuracy in the multi-target environment be increased by 18.6%.