| Tracked Land Robot is an intelligent mobile system, which can move autonomously and continuously on road or cross-country. Its research involves theories and technologies of multiple disciplines and reflects the latest achievements of information science and artificial intelligence. Due to its great significance in research and application, it receives high attention all over the world. Of all the key technologies, vision-based navigation is to perceive and understand the surrounding so as to find out a safe path for the robot. The road detection presented here is one of the key technologies in vision-based navigation.This paper primarily studies the road detecting. The process is as follows: image preprocessing, image segmentation, chain-code tracing, extracting feature points of road boundary etc. For the work of image preprocessing, a fuzzy enhanced method is designed based on HSI color space first. The proposed method can reduce the blur of road boundary and suppress noise of the road area, which improves performance.A new method is proposed for color image segmentation, which borrows the theory of color compensation. First, color compensation is carried out to original image, then blue channel is picked out and maximum entropy method is applied to blue channel image to obtain binary image. It can make system more robust. For binary image, chain-code tracing is used, and some sequent processing is applied to above tracing based on some prior knowledge so as to get the points of road boundary.The experimental results manifest that for the complex and varied road images, a better solution is provided in this paper, which guarantees the need of correct, real-time and robust for navigation and satisfies the practical requirement. |
PDF Full Text Request