We propose a mobile 3D reconstruction method for improving the precision and density of point clouds. It is suitable for hand-held scanners comprised of a colour camera and a lidar. We fuse time-synchronized and spatially registered images and lidar sweeps using deep learning techniques into dense scans, which are then used for progressive reconstruction in an odometry-like manner. We build a prototypic scanner and test our method in an indoor case-study. The results show that our pipeline outperforms reconstructions by other devices and methods, yielding relatively denser and detail-preserving point clouds with a 46% reduction in noise of reconstructed planar surfaces.