| abstract |
The invention discloses a semantically driven camera positioning and map reconstruction method, which belongs to the technical field of computer vision. The present invention first uses the semantic segmentation of the feature points of the current frame image; then according to the similarity and semantic category, uses the same kind of matching method to match all the feature points in the current frame and the key frame to obtain a matching pair; then through the current frame and the key frame All matching initializes the camera pose; then combined with the semantic judgment, the three-dimensional projection method is used to update the feature point matching pair; then the pose minimization is used to update all the feature point matching pairs; finally, the camera pose is used to construct a three-dimensional map; the present invention also realizes a A Semantics-Driven Camera Localization and Map Reconstruction System. The technical solution of the present invention not only performs multiple processes in the camera positioning stage, but also constrains the point cloud in the reconstruction stage, so that semantic segmentation, camera positioning and reconstruction systems are more closely combined, and more accurate positioning results and more perfect results are obtained. reconstruction effect. |