Research On Forward Vehicle Ranging Method Based On Binocular Vision

With the development of economy and the improvement of people’s living standards, there is a sharp increase in car ownership, and the traffic accidents are increasing year by year. Road traffic safety has become an issue with great concern of the whole society. Therefore the Advanced Driver Assistance System(ADAS) which can improve driving safety is widely recognized by the market. Vehicle detection and vehicle distance measurement are the important parts of the ADAS, which can alert potential dangers ahead and ensure safety. The front vehicle detection and vehicle distance measurement based on binocular vision are proposed in this thesis.In the vehicle detection part, to solve the problem of the sensitivity to illumination and the high false detection in the existing single method, a vehicle detection method based on the priori knowledge and machine learning is proposed. First of all, the region of interest is set and the enhanced image by the histogram equalization is generated based on the pre-processed image. Secondly, the vehicle candidate regions are extracted by using the MB-LBP feature and Adaboost classifier. Finally, the false detection is eliminated based on the horizontal edge and gray information of the candidate regions, and the vehicle detection is realized.In the distance measurement based on binocular vision, a vehicle distance measurement method is proposed based on the center-points feature matching of the vehicle rectangular frame in the left and right images. In the calculation process, the center-points of the vehicle in the left and right images are used to calculate the disparity, and the vehicle distance is calculated by using the principle of binocular measurement with the result of camera calibration.The experimental results show that the detection rate of preceding vehicle with the proposed method is more than 98.70% under the conditions of overcast, sunny and cloudy, and more than 89.64% at nightfall and nighttime. In the vehicle distance measurement based on binocular, the measurement error is less than 1% within 40m, and less than 3.6% in the range of 40m and 100m.