DEPTH RANGE ADAPTATION TO VARIABLE SCALE IN 3D-SCENARIOS FOR DENSE SLAM
This work deals with the problem of inadequacy of a fixed depth range in cost-volume-based dense methods for visual SLAM to fit different scales of the scenery. Mainly, the rapid changes of scale in the focused scene during navigation makes it necessary the adaptation of the depth range in order to maximize the resolution. We present a novel approach based on the distribution of the mass probability of the inverse depth estimations. We track continuously the wrapping of this function, keyframe by keyframe, and create corrections on the basis of soft restrictions to be effective in the next step. We illustrate the efficacy of the real-time approach employing a dataset of a scenario indoors.