Vision-Based Behavior Control of Autonomous Systems by Fuzzy Reasoning

Vision-based motion control of an autonomous vehicle operating in real world requires fast image processing and robustness with respect to noisy sensor readings and with respect to varying illumination conditions. In order to improve vehicle navigation performance in out-door environments, this paper presents methods for recognizing landmarks based on fuzzy reasoning. Firstly, a fuzzy thresholding algorithm is proposed to segment roads and to extract white line marks on streets from images. Secondly, some special domain knowledge about edges on roads represented by fuzzy sets is integrated into the rule base of a fuzzy edge detector. Based on this, the fuzzy thresholding algorithm is adopted to recognize road edges. In addition, a method for path planning in sensor space is presented and a path following behavior based on a fuzzy rule base is defined to control vehicle motion. The proposed methods are applied to navigate the THMR-III autonomous vehicle in out-door environments. Some experimental driving maneuvers have been performed to prove their effectiveness and their robustness.