Compressive sensing for reconstruction of 3D point clouds in smart systems

Performing an accurate 3D surface scan of everyday objects is sometimes difficult to achieve. Using the 3D scanner as a main sensor in a fast-moving mobile robot emphasizes this issue even further. When small robots with limited payload are considered, the professional Lidar systems are not likely to be embedded due to their weight, dimensions and/or high cost. Introduction of simple structured-light scanners makes possible fast scanning, effective robot detection and evasion of obstacles. Nevertheless, some obstacles may still be difficult to detect and recognize, primarily due to limitations of scanner's hardware which results in a low number of reconstructed surface points. In this paper a compressed sensing technique, primarily used for the reconstruction of 2D images, is utilized to enhance the quality of 3D scan, by increasing the number of reconstructed 3D points to the scanner's theoretical maximum. Obtained results demonstrated the feasibility of the approach in terms of mean square error.