Pedestrian detection using a LRF and a small omni-view camera for outdoor personal mobility robot

This paper describes a pedestrian detection method using a LRF and a small omni-view camera. In outdoor environment, the resolutions of LRFs are too low to recognize human reliably, and high resolution image requires high calculation cost for detecting walking persons. We propose a combination approach using these data. Particle filter based tracking and HOG (Histogram of Oriented Gradients) feature based identification enables to detect pedestrians with high reliability and effectiveness. Although the pedestrian identification from omni-view image needs computational effort for searching large area, LRF based tracking provides the identification process with limited interest regions in advance. We also reports how to construct a discriminative function which is able to cope with various resolution images. Finally, the proposed method was combined with gesture recognition and other robot system, and then an application to taxi service is introduced. The robot could find a person standing 15 meter away, recognized his swinging hand, and moved to the front of him.