Detection Of Preceding Vehicles Based On Edge Character

With the development of high speed transportation system in China, as a more and more important problem, traffic safety is attracting universal attentions of the society and the government. Freeway is oneway road. Surveys indicate that most malignant accidents are rear-ending collisions. So the development of intelligent transportation system (ITS) and intelligent vehicle (IV) is one effective measure to solve the traffic safety problem in our country, thus various research has been done on Safety Drive Assist System. One key point in the system is how to obtain the dynamic distance between two vehicles. So detection and recognition of preceding vehicles is a necessary prerequisite to measure the distance.At present, automobile anti-collision devices are mainly based on radar technology which can not be used widely because of its unstability and high cost. The information gathered by human beings, 75% of which is the form of images, is gained through the vision system based on eyes. Accordingly, we can also get what we need from images derived from other measure systems. Vision system has many advantages, such as wide detection range, integrity of target information and low cost. And it is more consistent with cognitive habits of people. Vision signals have irreplaceable advantages especially in detection of preceding vehicles.This paper put forward a new method of preceding vehicles detection based on computer vision. It adopts monocular CCD camera as an input device, and processes the input image by computer, then gains important characteristic data, finally complete detection of preceding vehicles by means of these characteristics. This paper is divided into three parts: The first part is edge enhancement and threshold segmentation of preceding vehicles under different illuminations. Thesecond part is extracting lines which have vertical edge features of vehicle. The third part is final location of the vehicle top edge.1. Edge detection and threshold segmentation is the foundation of extracting preceding vehicles'edges. According to image characteristics under different illumination conditions, this paper selects different edge detection and threshold segmentation methods.The paper adopts simple efficient edge detection algorithm to process an image under daytime illumination, then uses the maximum entropy principle based on two common threshold methods to select threshold to segment the enhanced image; under weak illumination conditions, the paper adopts improved edge detection algorithm, then selects another common threshold method to segment the enhanced image.2. Vehicle edges have been relatively clear afer edge enhancement and threshold segmentation. This paper figures out advantages and disadvantages of the traditional Hough transform and then uses a new detection method to extract lines which have vertical edge features.Considering the shortcomings of traditional Hough transform, the improved Hough transform was adopted to search lines that may be vehicle's edges in AOI (Area of Interest) which is founded based on lane mark knowledge. This method can extract three lines according to three maximum accumulator cells in AOI accurately and it also has a highlocation precision.3. Only two of the three lines mensionde above are vehicle's edges. The pseudo line can be removed based on location features of preceding vehicles in images. So the vehicle's vertical edges can be determined by the remainding two lines. Finally, the vehicle's top edges can be recognized and located on base of vertical ones.Based on three probable types of vehicle location characteristics in an image, true vertical edges of vehicle can be extracted and the upper edge was determined by vertical ones finally; As searching upper edges by the two vertical edges directly may result in inaccurate locations, the paper uses two vertical edges to locate anArea of Interest (AOI) and then verify identification of the upper level edges in the new AOI; In addition, detection precision and algorithm real-time performace have been taken into account, so this paper achieves rapid and accurate vehicle validating based on vehicle's rectangular AOI and NMI features.In a word, there is a much systematic and scientific research in this paper about detection of preceding vehicles under different illuminations. The paper adopts VC++ to develop image processing software and carries on some experiments to detect upper level edges of vehicle. Experiments'result achieves scheduled goal and proves that algorithms in the paper have good reliability and adaptability.