REAL-TIME MULTI-SENSORY DATA PROCESSING FOR PERCEPTUALLY RICH ENVIRONMENT DESCRIPTION

This paper describes methods that employ monocular vision in order to increase discriminative properties of the laser-based line segments typically used in SLAM. Visually salient objects are represented by clusters of photometric features and aggregated together with the basic line segments extracted from the 2D laser scanner data. An improved matching strategy for the visually salient features is introduced. Three recent photometric feature detectors are evaluated in the context of the proposed mapping approach, taking into account visual saliency, repeatability, and computing efficiency.