Fast Color-Based Object Recognition Independent of Position and Orientation

Small mobile robots typically have little on-board processing power for time-consuming vision algorithms. Here we show how they can quickly extract very dense yet highly useful information from color images. A single pass through all pixels of an image serves to segment it into color-dependent regions and to compactly represent it by a short list of the average hues, saturations and color intensities of its regions; all other information is discarded. Experiments with two image databases show that in 90 % of all cases the remaining information is sufficient for a simple weighted voting algorithm to recognize objects shown in query images, independently of position and orientation and partial occlusions.