Current students

3D Reconstruction represents a long-standing research topic in both Computer Vision and Robotics, used in different applications in the field of medicine, gaming, civil engineering, autonomous driving, tourism, etc. In the last decades, there was an important demand for 3D content for computer graphics, virtual reality and communication, further increasing the focus dedicated to the reconstruction problem to develop methods for accurate and fast results. Approaches in literature divide the reconstruction into multiple consequential steps: point based map estimation from a SLAM system, point cloud pre-processing, 3D reconstruction and surface refinement. In many real world situations, such as autonomous driving or microsurgery, this is not feasible because the reconstruction must be done simultaneously w.r.t. the localisation procedure. In this thesis we propose a multi-level real-time Mesh-SLAM system. The proposed system has three ascending levels of abstractions. The lowest layer consists of a SLAM systems, to perform localisation and mapping in real-time. At every key-frame, estimated cloud points and locations are used by the middle layer to perform a fast densification. This is sent to the highest layer, which performs surface reconstructions before the next key-frame is reached. The reconstruction layer also contributes, using the surface informations, to the adjustment of camera positions and point cloud estimation of the SLAM system, making Mesh-SLAM a fully dynamic simultaneous surface reconstruction, localisation and mapping system.