The Visual and Geometric Perception Laboratory (VGPL) was founded by Dr. Shen Cai in 2013, focusing on theoretical research and application technology in the fields of Computer Vision, Computer Graphics, and Artificial Intelligence, especially the 3D Vision tasks by using geometric and deep learning methods. The theoretical research directions of the laboratory include camera calibration, pose estimation, 3D reconstruction, robot navigation, feature extraction and matching, object detection, 3D object recognition, neural implicit representation, compression and reconstruction of 3D models, etc. In addition, the cooperation of companies also includes industrial vision inspection, action recognition, object segmentation, AR/VR, etc. There are currently 11 master students and 1 undergraduate interns in the laboratory. And 14 students and 6 interns have graduated from the lab.

Camera Calibration and Pose Estimation is a technical direction in Computer Vision to compute the intrinsic and extrinsic parameters of camera and distortion of lens. The research content includes calibration pattern design, feature extraction, homography estimation, distortion model selection and global parameters optimization. The theoretical research of camera calibration focuses on finding different geometric markers to form new set of constraint equations. The laboratory conducts theoretical and applied research on fast homography estimation, feature correspondence, calibration with conics, calibration with mixed primitives, rapid calibration of multiple cameras, , robot arm-based calibration, depth camera calibration, and zoom lens calibration. In this direction, the lab has published a number of academic papers and applied them to different company projects.

3D Object Representation/Recognition/Reconstruction is a technical direction in Computer Graphics based on known three-dimensional structure. Its research content includes three-dimensional representation, feature extraction, feature matching, classification, neural implicit SDF etc. It is one of the theoretical hotspots in the field of deep learning. The laboratory carried out theoretical research on multiple directions such as deep network based on spherical projection, sphere representation based on classification network, SN-Graph based on classification network, and 3D object coding based on key spheres etc., and published several academic papers.

Image based 3D Reconstruction is a technical direction to reconstruct the scene or the three-dimensional details of an object. Its research content includes feature extraction, pose estimation, three-dimensional representation, extrinsic parameters and joint optimization of three-dimensional points. The laboratory has explored multiple directions such as depth camera based fusion reconstruction, multi-view stereo, and reconstruction with calibration patterns, and applied them to different company projects.

Primitive extraction and correspondence is a technical direction of extracting geometric primitives such as points and lines from images. Its research contents include feature point extraction, descriptor extraction, line extraction, primitive correspondence, and region or image matching. The laboratory has carried out theoretical and applied research in multiple directions such as fast outlier removal, feature descriptor extraction, and geometry-based correspondence.