Abstract: Aiming at the problem of pose estimation in 3D texture reconstruction of cultural relic image, this paper proposes a high-precision pose estimation method based on 3D data-driven and adjustment. Firstly, the accurate homonymous features are obtained through image feature extraction, matching and false matching point elimination. Then the high-precision pose relative relationship of the first pair of images is established, and the relative pose of the subsequent images is determined by combining the two-dimensional index driven by the 3D point data of the first pair of images. Finally, the high-precision image pose parameters are determined by beam adjustment and LM algorithm. The experimental results show that this method has certain advantages in determining the pose speed and pose accuracy, which lays the foundation for the subsequent generation of dense point cloud.

Key words: sequence image, image matching, data driven, image pose, high-precision