Vision-based Preceding Vehicle Detection In Highway

With rapid development of electric technology, research of automotive intelligent driving assistant has become a focus. The vision-based collision warning system is an assistant system that is to give driver warning through sound or light signal by apperceive latency danger of traffic circumstance, and can help to make up with the shortage of driver and improve the safety of driving by avoiding traffic accident.Vision-based driving condition perception system plays an important role in intelligent vehicle, for it has a certain of advantages that other perception system can't compare with, for example low cost, ample data providing, non-pollution to the environment etc.Vision-based vehicle detection utilizes the camera installed in the car to get images real time. Then the outline of the preceding car can be extracted from the images by image processing and segmenting. The relative distance between two cars can also be calculated by coordinate transform. So the system can be used to provide information for distance keeping and collision precaution system.According to the whole theory and algorithm of the paper, we have several conclusions as below:Firstly, by study and analysis of intelligent vehicle system developed in the world, especial for the vision-based perception system, a basic understanding about intelligent vehicle framework is made. Comparing withthe different method applied in the different system, advantages and disadvantages are summarized. Learning from the experience and according to the condition of our library, this article put forward a new method.Secondly, by image processing and segmenting, the preceding vehicle is outlined with a red rectangle. First, the average pixel value of road area is sampled by a statistical method. The road and non-road area can be separated. So the noise in non-road area can be eliminated by this method, which is favorable for vehicle detection. Once the approximate location of the vehicle is calculated, it can be used as a sub-ROI for further detection. An accurate location can be calculated by shade dilation algorithm and coalition.Finally, as a basic knowledge, there is no big difference during a serial of frames captured within a second, so there are a lot of resemblances. For example, a vehicle can not disappear at once if it has appeared in the last frame. So this paper utilizes the correlation of the serial frames. The detection is robust and real-time by using the enhancement method mentioned above.The creative work of this paper is: The algorithm in this paper can detect vehicles in both self-lane and lane beside via coordinate calibration. Meanwhile, the application of correlation of serial frames makes the detection more accurate and robust.