Support system image understanding for navigation

We have developed an navigation support system for the blind. The system can recognize the rough shape of objects quickly in office scene, and can represent visual information linguistically to the blind. For the image understanding system used in the navigation support system, how fast the system can offer the visual information in a dynamic environment becomes extremely important in such an environment where one cannot limit the position of the camera, its distance and direction to the object. In this system, we first apply a quick extraction of prospective object regions from color images using the physical features as a low level knowledge. Then we propose an approach to infer the regions synthetically with high level human knowledge using fuzzy logic and Dempster-Shafer´s theory. The system represents the inference result linguistically according to the degree of its authenticity