Dynamic targets detection for robotic applications using panoramic vision system

Abstract

This paper presents experiments in dynamic targets detection using panoramic images, which represent rich visual sources of global scenes around a robot. Moving targets (people) are distinguished as foreground pixels in binary images detected using a modified optical flow approach where the intensity of lighting source is variable. The directions of detected targets are determined using two strategies; the first one is convenient for unfolded panoramic images; it searches most probable regions in the last binary image by calculating a histogram of foreground pixels on its columns. The second approach is applied on raw panoramic images; it regroups foreground pixels using a technique that generates a new pixel's intensity depending on the intensities of its neighbors. © 2008 Springer-Verlag Berlin Heidelberg.