Detection Of Preceding Vehicles At Night Based On Infrared CCD

In recent years, with the development of national economy and the rapid increase of auto possessive quantity, traffic jam and accident has become an issue that cannot be ignored. According to some statistic data, rear-end collisions cased by a variety of reasons occupy a large proportion of traffic accidents,and rear-end collisions are mainly caused by lack of concentration. Therefore, the forward collision warning system should be researched. This system can remind the drivers timely, and it can avoid many rear-end accidents. One key point in this system is how to obtain the dynamic distance between two vehicles. So detection and identifier of preceding vehicles is a necessary prerequisite to measure the distance.At present, detection and tracking of preceding vehicles during the day has been researched in-depth in our group. Nowadays, the algorithm library of our group can meet the requirements of vehicle detection and tracking during the daytime. And the forward collision early-warning system can work effectively in the day.Night driving is unavoidable. Because of poor visibility at night, driver fatigue and other factors, night driving is more dangerous than driving in the day. Some related statistics indicate that the rate of night-time road traffic accidents are twice more than that during the day, and vehicle rear-end collisions account for one third of the entire accidents. So the research of the identification of preceding vehicles at night has very important significance to the safety of driving at night.Detection and tracking of preceding vehicles at night has been studied preliminaryly in our group. The algorithm mainly used preceding vehicles taillights to detect and identify the preceding vehicles. This algorithm is not adapted to all conditions at night. When the taillights are dark or not light, the tail of the vehicle looks like a whole part, taillights are connected, the algorithm is easier to fail and the preceding vehicle cannot be identified. When the taillights are not light and the environment visibility is poor, the preceding vehicle images we grab with the ordinary CCD are not ideal. The taillights, even the vehicles outline are not clear. In view of this situation, Infraed CCD is used to grab preceding vehicles images instead of ordinary CCD in this paper. In this paper the image acquisition system includes infrared CCD and infrared source. These images are processed by computer, and then gains important characteristic data, finally complete detection of preceding vehicles using these characteristics.This paper is divided into three parts: The first part is analyzing the preceding vehicles' infrared images, and dividing these images into it two categories according to the preceding vehicles' characteristics, and then preprocessing the infrared imagery. The second part is threshold segment of preceding vehicles images with obvious taillights and license plate, extracting circularity, rectangularity, height to width aspect ratio and symmetry, and identifying the preceding vehicles. The third part is edge enhancement and threshold segmentation of preceding vehicles images that taillights and license plate are adhensive, extracting the vertical edge symmetry, height to width aspect ratio, the gradation symmetry, and recognizing the preceding vehicles.1. Research of Infrared image preprocessing method. According to analysis, infrared image has some characteristics, such as low signal to noise ratio and contrast. In consideration of this, this paper selects median filter to remove noise, and choose grey linear transformation to enhance contrast.This paper puts forword two different methods to detec preceding vehicles. One is using taillights and license plate, the other is using the vertical edges of vehicles.2. Preceding vehicles detection at night based on taillights and license plate. This paper analyzes preceding vehicles images and finds that most of them have obvious taillights and license plate when the taillights are not light and the environment visibility is poor. This part compares moment preserving thresholding method with segmentation method based on fuzzy theory, and chooses segmentation method based on fuzzy theory.In order to save time and detect much more accurately, this method extracts area, circularity, rectangularity, height to width aspect ratio and symmetry in AOI (Area of Interest) which is founded based on lane mark knowledge. Finally, relizes vehicles detection.3. Preceding vehicles detection at night based on edge character. When taillights and license plate are adhensive, this paper uses sobel vertical operator to enhance vertical edges and choose moment preserving thresholding to threshold images.This mehtod uses vehicles vertical edge symmetrical characteristic to determine the symmetric axis of AOI. And then seeks for vehicles' left and right edges. At the same time, a new AOI is founded based on vehicles' left and right edges. In this new AOI, the upper and lower edges are determined by gray mutation between the vehicles and background. Then this paper extracts gray-scale symmetry and height to width aspect ratio in the new AOI. At last, preceding vehicles detect and identify based on edge symmetry, gray-scale symmetry and height to width aspect ratio.In brief, detection of preceding vehicles at nignt on the highway has been systematically and scientifically studied. The paper adopts Visual C++ to develop image processing software and carries on some experiments to detect and identify preceding vehicles at night. Experiments'result achieves scheduled goal and proves the glgorithms in the paper have good reliability and adaptability.