Image Linear Feature Extraction Used In Road Detection Of Vision Navigation

Intelligent Vehicle has many functions such as environment apperceiving, planning and decision-making, multiple-level driving assistance and so on and it is an integrated system. Because of its promising application value, IV has received considerable attention of so many scholars from its coming into being and made the wave of IV research reach all over the world. Automatic navigation acts as the basic function of IV. Furthermore, it is the key technique of IV. Most of the IV systems acquire guiding information by multi-sensor, which include computer vision, radar, GPS receiver, infrared detector and laser detector etc. Computer vision is based on the model of human eyes and it can explain the fuzzy visual environment, so it becomes the main navigational means of IV. The technique of road edge detecting is a very important technique and must be solved intelligently. It also presents many potential applications. For instance, it can be used in Drive Assistance System and Drive Warning System to develop the active safety of vehicle, the efficiency of transportation and the driving comfort. It also can be used to realize the automatic pilot for the standard road vehicle. The lane marker acts an important role in many IV systems based on vision navigation to realize the function of automatic navigation, such as in Vormas-p and Navlab-5. But it becomes very difficult to be detected for computer vision when the lane marker is contaminated, or covered, or deformed. To improve the flexibility and adaptability of the automatic navigation of IV, the paper has provided the navigation information for IV through detecting the nature road edge. Meanwhile, the method can also detect the lane marker. Because the nature road edge will not be easily polluted like the lane marker does, so it indicates better adaptability and better ability to be immune to the disturbance. It is very important for automatic navigation to detect the road edge correctly, fast and steadily. The research of the paper is about the method to extract the linear feature of the nature road edges in the navigation pictures. To improve the efficiency and adaptability, an algorithm is developed. The algorithm is based on the linear road model and it includes two stages: initialization detection and tracing detection. The pixel feature and the frame feature are the basis to identify the road edges, work respectively in initialization detection and tracing detection. The linear road model in the paper makes the algorithm extract the feature from the image with less computer storage, less computation complexity. The paper is based on the model of human eyes to analyze the image in a global perspective, which helps to resist to the disturbance of the redundant information in the image and helps to recognize the edges more efficiently. Although most roads are not linear as an entity, the curvatures of the road are little and it is reasonable to use the linear road model to detect the roads in most urban traffic conditions. In the paper, the edges of the road are supposed to be in the same ground plane and the width of the road is supposed to be invariable. The road constrains above ensure the linear road model works. In the paper, dynamic region of interest algorithm works in the tracing detection. In the image for vision navigation, most part of the image is useless to detect the road edges except the region around the real edges of the road. The region of interest in this paper refers to the region with useful image information, which is around the edges of the road. The tracing detection only proceeds in the region of interest, which limits the detecting area and save much time. In the initialization detection, enhance the image first, then extract the edges of the image with Sobel operator, after that, extract the linear feature of the binary edge image with Hough transform. Some constraints will make the Hough Transform extract the linear feature more efficiently, such as: commonly, the left edge of the road is in the left half plane and the right edge is in the right half plane, if the beginning position of the car is in the middle of the road, the obliquity of the left edge is between 0 and π/2, the obliquity of the right edge is between π/2andπ. The constraints above will limit the search area of Hough transform and save much time. After the recognition of the feature of the edges, if the road deviation, which is calculated from the two neighbor images, is less than the appointed value, it is assumed that the vehicle is tracking the edges detected steadily. Then the algorithm turns to the tracking detection. In the tracing detection, the...