Fast people detection in indoor environments using a mobile robot with a 2D laser scanner

摘要

The high uncertain and dynamic nature of indoor environments, especially including dynamic targets or obstacles, makes the realization of autonomous navigation and optimal action generation of indoor robots a challenging task. In this paper, by combining a leg pattern detection strategy and an AdaBoost based online learning method, a fast multi-level detection approach has been proposed to solve unknown people identification and extraction problem using an indoor mobile robot with a single 2D laser scanner. The laser data is firstly preprocessing to detect vertical edges. According to the classification of three typical leg postures, leg patterns are extracted from vertical edges to achieve the initial detection of the human body. After that, the AdaBoost based online supervised learning method is used to construct a strong classifier for further classification of above ambiguous patterns with two legs together or only a single leg, which can improved the detection accuracy of dynamic people. Experimental results show the correctness and effectiveness of the proposed approach.