AUTOMATIC SCALING IN 3D MAP BUILDING FOR SLAM
Mario Jordán, Emanuel Trabes
In this work we develop a novel approach for indirectly estimation of the metric scale for dense mapping of 3D environments. The scaling-factor emerges as a local estimation for providing in real time a metric scale to the depth map. The approach is suitable for real-time monocular SLAM applications. It employs a laser arrangement fixed to the camera whose beams commonly
impinge on middle-distant 3D points that are tracked in the frame sequence. The method employs a so-called wildcard frame and a keyframe to estimate the scaling factor along with the optimization of an energy functional to provide good depth estimations at middle distance. Purpose-built experiments are led to illustrate the approach performance in mapping and tracking and show its feasibility.