Study On The Method Of Recognizing Shadow Navigation Lane For CyberCar

Along with the increment of the automobile amount that many countries preserves, many social problems come out, such as fatal petroleum crisis, traffic justle, traffic jam and the security of people because of the extend city transportation, serious environment pollution, astonishing automobile garbage and its recovers etc., all these have lowered the city life quantity. In order to resolving this problem effectively, a kind of intelligent transportation system-CyberCar System emerges with the tide of times. Among so many kinds of region intelligent vehicles, vision-based intelligent vehicle makes many countries concentrate on it because of its abundant information, widely used and the special superior intelligence, and become the main researching field and development direction on the vehicle guidance aspect. But because the outdoor environment especially the shadow environment if erratically changing and can not be predicted, the lane recognition of the vision-based intelligent vehicle becomes difficult. Till now, there are few people who have devoted to research the recognition algorithm of the shadow lane in different illumination degree, so how to improve the reliability of the shadow lane recognition algorithm becomes the researching focus. In this paper, algorithms of recognizing navigation lane are studied for vision-based CyberCar in most kinds of shadow environment of different illumination degree according to the image characters of the corresponding kind and in view of the actual application, and technologies such as artificial neural network (ANN), digital image processing and fuzzy logic theory are used. (1)Method of segmenting the shadow navigation lane in faintish illumination is studied by using threshold surface. Original points for the threshold surface are selected using the grads value according to the characters of this kind of image, and get rid of the false points according to the grads direction histogram. The threshold surface is formed by the matrix operation using the edge points of different position as input. This surface matches the gray changing trend in the original portrait direction, so it has a good approach to original image. Examination of this kind of images segmentation including different tree's shadow, passenger's shadow, car's shadow and building's shadow shows that the algorithm is effective, and the algorithm takes 125ms/frame. In order to improve the real time ability, this paper handles only one row every two rows, then the points for segmentation reduce to a quarter of the image, and handles the other frames in the area of interest after the first frame segmentation, so the time reduces to 47ms/frame, satisfies the real time required. (2) Algorithm of segmenting the shadow navigation lane in normal illumination is studied by using region growing method. Researching on selecting the seeds, establishing the growing rules and confirming the time when growing is over is carried out according to the theory of the region growing algorithm. Original image is divided to ten blocks and segments every block using the region growing algorithm in order to avoid the phenomenon of gray asymmetry. The seed points'selection is very important and grads vector is used to the seed points'selection. The information of the lane width is used to get rid of the false points after getting the original seed points. During the growing course, whether the next point is grew or not is decided by the grow rules which using the combination of the gray average and the gray equation of the pixels in the area that has grew. The image segmentation finishes when all the points that fit the rules have grew. The algorithm time is 47ms/frame by segment one row every two, and this time is enough for the real time system. Segmentation examination of this kind of images including different tree's shadow, passenger's shadow, car's shadow and building's shadow shows that the algorithm is effective. (3) Method of segmenting the shadow navigation lane in strong illumination is studied by using Laplacian edge detection algorithm. First calculates the grads value of all points in the image using the Laplacian algorithm. According to the characters of this kind of image, the grads value of the points in the lane area is zero and that of the points in the background is none zero, so the object area and the background area is divided. After that, uses the plumb projection of the white point amount to get rid of the less isolated yawp points. Segmenting method using...