Study On Lane Detection And Vehicle Recognition Based On Vision

In recent years, with the progress of society and development of road transport,traffic safety has become increasingly important social issue, Intelligent Transportation Systems(ITS) have emerged and received wide attention. Intelligent vehicle(IV) is an important part of ITS, it integrates the functions of environmental perception, decision support, and automatic driving, represents the future direction of the vehicle, has a high academic value and broad prospects for engineering applications.In this paper, under engineering background of Beijing University of Technology electric Smart car BJUT-SHEV, studied key technologies of IV based on monocular vision —— Lane and vehicle recognition. The main contents include the following four aspects: image pre-processing, lane detection, vehicle identification, coordinate transformation and parameter calibration.Firstly, the image pre-processing technology is studied. Gray-scale image obtained by combining the weighted average method and the color channel extraction method. Reduce the noise information of gray images using the median filter in 33?templates. Classify images based on light intensity,then enhance the gray scale information using histogram equalization method. Contrasts optimal threshold method and OTSU, choose the latter for adaptive segmentation.Secondly, two different traffic lane detection algorithms are proposed and modified, carries on the contrast analysis. The method of progressive retrieval by screening feature of lane points is proposed. Take the advantage of road image characteristics, the upper bound of processing region and the true road area are divided. Least Squares is used to fit feature points, so that get the actual parameters of lane as to reconstruct the road model, then reference test results to determine the direction of the road. An improved Hough transform method is proposed. Connected region labeling method is used to filter background and noise information. A custom difference operator is designed for detecting lane edge. Dynamic area of polar angle and polar radius is established, improve the speed and stability of the algorithm.Based on the recognition result, design departure warning method for safe driving assistant. The progressive detection method has better real-time performance, and it can provide more effective information during the cornering. Improved Hough Transform has stronger robustness and anti-jamming capability.Thirdly, vehicle recognition method is proposed. after the improved Hough transform, road gray area and Secondary OTSU are combined together in order to identify the shadow features under vehicle. Through the corrosion and expansion method, the interference points are filtered out, then merge the shadow lines, extractthe region of interest(ROI). ROI will be selected and identified by using entropy and symmetry, these methods can reduce the missed or false detection rate. An improved method is proposed for Robinson direction operator, application of Kalman filter algorithm to predict the position of the vehicle, achieved good results in extracting boundary of vehicle.Finally, the conversion relation between camera image coordinate and intelligent vehicle coordinate is established. Obtain relative velocity according to test results and refresh rate. Make instructions for intelligent vehicle based on the distance and relative speed, converts it to the data that planning decision computer understand it directly.In order to verify the algorithms, a experiment platform is developed, a lot of experiments about the key technologies have been done. The results show that the system has good real-time performance and anti-jamming capability in different road and image features.