The problem of 3D point cloud and 2D image pose error of mobile mapping systems has always been a prominent topic, and a new panoramic image pose optimization method based on point cloud was proposed in this paper. First, 3D point cloud are projected to a depth panoramic image by image feature enhancement including masked discriminant projection, which could modify the image quality. Then a Siamese neural network with ResNet50 feature extractor is employed to extract projected image features. The results could be used to improve the rotation matrix errors between the panoramic image and the point cloud. The essence of the new proposed method is to reduce the relative error between the LiDAR and the camera, and to provide a good pose for the camera, which could modify the matching accuracy between the 3D point cloud and the 2D image. Experimental results show that the method studied in this paper can significantly improve the localization performance. The completion of this work helps to improve the accuracy and reliability of the mobile measurement scanning results data, which is important for point cloud coloring, object segmentation, map construction and navigation.