A fuzzy approach in color object detection

Probing over a digitized image of robots taken from a top view to uniquely identify them is not an easy task. The recognition process, which involves scanning a digitized image and characterizing it, is made difficult by varying illumination, position and rotation. Furthermore, the vision system is plagued with inherent difficulties that cannot be completely controlled. Effects such as lighting and shadows, lens focus, and even quantum electrical effects in the sensor chip combine to make it essentially impossible to guarantee that the color being tracked would remain constant as the robot traverses the exploration field. From among the different recognition cues, like shape, size, position, and motion, this research focuses on color as the primary discriminating feature. However, as color is greatly affected by so many underlying factors, fuzziness is incorporated into the system to address the problem of uncertainties in color object classifications. The computing potential of fuzzy logic in the field of machine vision is very promising, but not yet fully explored. The paper presents an approach that combines fuzzy logic with graph-theoretical clustering techniques in order to add flexibility in defining object colors, and to recognize robots fast and accurately. The recognition scheme is primarily divided into three subtasks: feature extraction, fuzzy system configuration, and object identification. Algorithms are described for each subtask.