Obstacle Detection using Structured Background

The paper presents an obstacle detection method for mobile robots using a structured background. This method is based on differences of appearance between obstacles and the background in the camera images. The basic idea is to cover the ground of the workspace with a known structure. If part of this structure is obscured, this can be detected and indicates the existence of an obstacle. The method uses a reference symbol, chosen to exhibit certain features, to construct a reference image of the structured background. To detect possible obstacles we calculate the Fourier descriptors (FD) of all contours included in a given image and compare them with those of the stored reference symbol. This enables us to recognize all reference symbols which are not obscured by obstacles. We then determine the positions and dimensions of all existing obstacles by calculating the occupied symbol areas. The method is implemented as part of a robotvision system for fully automated stockkeeping. In this paper the results are shown using a MATLAB implementation.