To improve the accuracy of geometric modeling of the telerobot virtual environment, a method for correcting the geometric modeling of the telerobot virtual environment based on the visual-tactile fusion is proposed. Firstly, based on the visual information, the geometric modeling of the target is realized through point cloud acquisition and preprocessing, point cloud registration combined with thick and thin registration, surface reconstruction and pose measurement. Then, aiming at the application background of telerobot, a dexterous hand control mode combining local semi autonomy and multi finger cooperation is designed to complete the collection of target tactile point cloud. Finally, the visual-tactile fusion method is used to fuse the visual point cloud and the tactile point cloud through KDtree, and the visual-tactile fusion point cloud is used to modify the geometric model. Experiments show that the geometric dimension error of the geometric modeling of the target with different materials by this method is less than 3. 6 mm, the positioning error is less than 6. 8 mm, and the attitude angle measurement error is less than 4. 3°. The effectiveness is better than the geometric model based on visual information and the geometric model based on tactile information. The geometric model constructed by this method not only includes the color information of the target, but also improves the accuracy of the geometric model. It performs well in the details of the model, which effectively combines the advantages of visual and tactile modal information.