Multi-view underwater 3D reconstruction using a stripe laser light and an eye-in-hand camera

Autonomous manipulation in unestructured underwater scenarios is a high challenging skill that has been poorly studied and is becoming more and more important in the last years. One of the main problems regarding the autonomous manipulation, is to find out the characteristics of the object which is going to be manipulated. This paper presents a new approach to obtain an accurate 3D reconstruction of this object. This approach consists in attaching a laser stripe emitter and a camera in the forearm of a robotic arm. Moving the arm, the laser scans the scene where the object is and, at the same time, the camera records the scan. Thanks to the arm and the position of the camera, the scene can be reconstructed from different views and from a position close to the object. The recorded images are processed to obtain the 3D position of the part of the scene projected by the laser. Before the intervention, a process of calibration is needed to calculate the relationship between each part of the system. Furthemore, in order to reduce the time of processing of the images recorded during the scan, an optimization algorithm is presented which consists in discarding, before the processing, the pixels of the image which do not contain relevant information. The approach herein presented and the optimization algorithm are tested using an underwater simulator.