Study On The Modified Vision Algorithm Of Lane Departure Warning System

With fast development of express highway and gradual increasing of vehicle population, the incidence rate of road traffic accidents, especially the heavy accidents, is increasing. So the problems about traffic safety have been increasingly prominent. In this case, how to improve the safety performance of vehicles and decrease the road traffic accidents has become human's focal point and researchers'puzzle. Surveys indicate that many of traffic accidents happened in express highway are due to drivers'fatigue or their attention not concentrate so that lane departure or alter.Based on this kind of reason, lane departure warning system emerged. It could assist driver to keep the vehicle running in the lane. Its operating principle is like this: a camera is installed on the vehicle to capture the picture of front road, then system can get the information of vehicle in current lane by processing. As soon as it detects the vehicle is near the lane and the driver doesn't turn on turning lamp, it will warn the driver correct the unconscious lane departure in order to avoid the occurrence of accidents.This paper researched on the algorithm of vision system for lane departure warning system on the foundation of original algorithm and hardware platform based on the systemic and profound study on the algorithm of lane departure warning system developed successfully by intelligent vehicle group, as well as some long-distance road test. The purpose is to improve and extend original algorithm, so that the system could be input to market as soon as possible, display its advantage on the safety assist.The content of this paper is divided into four parts: the adjustment of camera's internal parameter based on the feature area of images, the detection of lane marker based on the improved Hough Transform, the useful information capture of lane and the calibration of camera as well as the construction of road model.1. From the view of market, the cost of this system is the main consideration for producers and Transportation Company, which is the prime reason why our native system could not be input to market.Intelligent Vehicle Group of Jilin University has finished the integration of hardware, and the system's cost can be accepted by most of users. Based on the hardware device, the character of camera is studied, combined with its application background, the camera's gain, light and exposure time can be adjusted according to the feature area's information of image .The feature area is fixed by the program's state: detecting or tracking the lane marker. Test proves that the technology can increase the contrast between lane marker and the other section of road, so it can decrease the time drain of algorithm, correspondingly reduce the demand for computer's processing system.2. Sometimes when the vehicle is deviating or changing the lane in the test, the recognized lane larkers is not the current ones, which would cause the system not display its alarm function. Studying the advantages and disadvantages of traditional Hough Transform, the article uses improved Hough Transform to detect the straight line, then extracts the current lane marker according to some bounded terms. The improved Hough Transform first selects the ranking of each lane marker, then takes the Hough Transform and finds their maximum value. Compared with traditional Hough Transform, the improved method does more adapt to the situation existing multi-lane marker.3. Studied the camera's mounting attitude is how to influent the algorithm, then in order to get useful road information, the paper adopts the effective field of vision to mount camera, which could capture the current lane marker's information, and also keep the redundant information from being captured in order to decrease the processor's treating time. 4. In the original algorithm the model is built on the basis of camera's horizontal mounting attitude and fixed mounting height, but in the factual application, it's hard to make the camera mounted horizontally, and it is a trouble to measure the mounting height. So form the view of practicability, the paper employs a kind of calibration model. As long as we know the image and world coordinate of calibration dots, then the image and world coordinate can be connected. The test data shows that this method could satisfy the system's demand, and the calibration course is simply, so the practicability of the system is improved.5. In the original algorithm the current running parameters are got on the base of camera's horizontal mounting. But when adopting the effective field of vision to mount camera, a suitable model need to be established to get the vehicle's running parameters. By analyzing the advantages and disadvantages of most lane departure warning system's models, the article confirms vehicle's location in the picture, then by estimating the position of pre-aim point, transforms the parameters in the image to ones on actual road. Long-distance test indicates that the method could keep the lane departure warning system's accurateness.In brief, this paper researches some algorithms of the vision system based on the lane departure warning system developed successfully by our group, and the modified algorithm improves the system's real-time performance, accuracy and practicability. A large amount of actual long-distance proves that improved lane departure warning system has the qualification to become product.