Study On The Algorithm Of Recognizing Navigation Lane Based On The Road Edge For CyberCar

The area intelligent vehicle (CyberCar) is a kind of low-speed electric intelligent vehicle, it is the outcome when people transfer the research to the outside area traffic from resolving the production line's automation, and it is a new direction in intelligent vehicle field. It depends on highly developed modern computer technology, information technology, electronic technology, artificial intelligence and network communication etc., and can be dispatched and managed by the control centre. At present, CyberCar is only applied some special areas, such as large playing place, exhibition center, public touring area and airport station etc., and resolves communication problem in these area. It is free of noise and pollution with the characteristic of high safety and convenience, and reflects the idea of"People Is First"in traffic behavior. With higher demand of traffic safety and environmental protection, CyberCar will be put into operation in more fields. It will eventually become an important symbol with which social civilization and technical advancement is measured.In recent years , vision navigation become one of the hotspots in self-navigation fields because of computer image processing technology. The Intelligent Vehicle Group of Jilin University has successfully developed experimentation vehicle, JLUIV-5 CyberCar, which can recognize two-dimensional white strip path paved on the ground by computer vision, it has realized the straight line running, large turning, increasing and decreasing speed automatically, parking automatically, parking to avoid barrier, and so on; but it also faces a lot of problems in the process of experimentation and practical application: First of all, the long distant road surface experimentation needs to lay long white strip as navigation path, it takes too much time and the cost is extremely high; Furthermore, laying the white strip has been restricted in practical application. So it is difficult to carry out the practical experimentation. The thesis tries to take the natural road border as the navigation path of JLUIV-5 CyberCar, study how to recognize and track the natural road border in various illumination reliably and steadily.The thesis briefly classifies the road border images three types, which are collected in complicated illumination by CCD camera: the first is road images in regular illumination; the second is in weak illumination; the third is in strong illumination. Indeed, this classification is a simple partition of many static sample image collected by CCD based on people's vision, the purpose is to study the effective algorithm of recognizing and tracking the road border according to the image characteristic under different illumination condition.Using a method based on Zernike Moment according to the characteristic of road border images under regular illumination, we can get road border feature points and road border parameters, and then track the road border by building the echelon area of interest, the essence of the track is to contract the area of image processing. Besides, for the limit of installing CCD, the thesis just recognizes the right border of natural road.For the road border image under improper illumination, the thesis classifies the images weak illumination image and strong illumination image. For the image under weak illumination, enhancing the road border by Sobel operator, then choosing an overall threshold to segment enhanced image. The threshold is selected by the invariability of image moment; For the image in strong illumination, enhancing the road border by using multi-scale wavelet, then choosing a threshold to get binary image by iterative method. The binary images pass filter processing to decrease the disturbing points. Finally, the road border in two kinds of images is located by Hough Transform, and is tracked by the method based on Hough Transform. The essence of the track is point-tracking in Hough space. Because of the limit of installing CCD , the right road border is recognized only.By analyzing the characteristic of plentiful sample images and evaluating the validity of recognizing road border, the thesis at last has designed an algorithm classification, which is based on part average value and part variance, thereby the system can classify the primitive road border collected by CCD according to the part average value and part variance in special window, then sorted images are processed by the corresponding algorithm and finally locate road border and gain parameter information of road border. Because of the complex illumination outside, the thesis sets up dynamically parameters of CCD exposure time, brightness and gain according to the image part average value to avoid brighter or darker road border image collected by CCD. This regulation is equal to the function of digital camera automatic focusing and automatic exposure. Dynamic experimental results prove setting up the CCD exposure time parameter and other parameters can effectively restrain CCD to collect brighter or darker road border image, raising the reliability of entire system recognizing the road border.The software application system of recognizing and tracking road border is developed by Visual C++, takes the JLUIV-5 CyberCar as the experimental platform, carries out lots of road experiments in Nanling Campus of Jilin University. The experimental results indicate that , the road border recognizing system has good acclimatization, the algorithm takes 60ms averagely, the validity of the road border recognizing can reach 95%, it has satisfied the navigation demand of JLUIV-5 CyberCar fundamentally.To sum up, the thesis has done system research of the non-structure road border deeply, and has realized the CyberCar automatic navigation by recognizing natural road border. The research results indicate that, adopting machine vision and taking the natural road border as the navigating path of low speed manless driving vehicle are completely feasible. Its achievement has certain reference value toward different information amalgamation in the field of intelligent vehicle navigating.