Camera Pose Optimization for 3D Mapping

Digital 3D models of environments are of great value in many applications, but the algorithms that build them autonomously are computationally expensive and require a considerable amount of time to perform this task. In this work, we present an active simultaneous localisation and mapping system that optimises the pose of the sensor for the 3D reconstruction of an environment, while a 2D Rapidly-Exploring Random Tree algorithm controls the motion of the mobile platform for the ground exploration strategy. Our objective is to obtain a 3D map comparable to that obtained using a complete 3D approach in a time interval of the same order of magnitude of a 2D exploration algorithm. The optimisation is performed using a ray-tracing technique from a set of candidate poses based on an uncertainty octree built during exploration, whose values are calculated according to where they have been viewed from. The system is tested in diverse simulated environments and compared with two different exploration methods from the literature, one based on 2D and another one that considers the complete 3D space. Experiments show that combining our algorithm with a 2D exploration method, the 3D map obtained is comparable in quality to that obtained with a pure 3D exploration procedure, but demanding less time.