Panoramic feature matching method based on non-Euclidean space

The invention discloses a panoramic feature matching method based on a non-Euclidean space, which comprises the following steps of: (1) reading two panoramic images to be matched in an equidistant projection form, and respectively mapping pixels of the two panoramic images to a regular icosahedron grid subdivided for eight times by using nearest neighbor interpolation according to a panoramic camera imaging model; (2) extracting a feature map by using a feature pyramid network based on spherical convolution; (3) performing rough matching and fine matching on the feature maps of different levels by using violent matching; and (4) finally, removing mismatching by using a random sampling consistency method. According to the method, feature matching can be directly carried out in the non-Euclidean plane, the distortion influence of plane projection on the panoramic image is avoided, and the panoramic feature matching precision is effectively improved. 本发明公开了一种基于非欧式空间的全景特征匹配方法，包括以下步骤：(1)读取待匹配的两幅等距形投影形式的全景图像，根据全景相机