The Research Of Vehicle Forward Collision Warning System Based On Vision

The vehicle front collision warning (V-FCW) system based on vision is one of the most typical applications in the field of automobile safety driving technology. FCW mainly uses the sensor to obtain information, then processes it, to accurately determine the risk of collision in front of the possible situation and to give warning to the driver. V-FCW system based on vision uses camera as an environmental sensing tool for obtaining information, which is different from the current related research of using radar sensors in V-FCW system. The vision sensor has the advantages of pattern recognition and lane detection, but at the same time, it also faces the huge technical challenges, such as obstacle recognition and ranging. Based on this research background, this paper designs a set of V-FCW system, in which a new method of vehicle detection and vehicle distance measurement, besides a safe distance calculation algorithm, are proposed. Its effectiveness was verified through simulation analysis.On the basis of the analysis of vehicle front collision warning safety distance model, according to the running state of the vehicle and braking characteristics, this paper establishes a new car safety distance model, explaines and analyses parameters of the model, with fullly considering the brake of a vehicle driving factors and environment factors. Combined with other model algorithm, MATLAB was used for simulation analysis. According to the change rules of the minimum safety distance with the velocity and the adhesion coefficient under different conditions, it was proved that the model could better reflect the real running state of vehicle. What was more, the large deviation of the calculated safety vehicle distance was reduced.This paper also makes some research on vehicle detection and recognition. Class Haar and improved AdaBoost algorithm for the detection of the front vehicle were used. Firstly, the extended class Haar feature of image was calculated based on integral graph; Then the extracted class Haar feature value vector set was used to improve the AdaBoost algorithm. Improvement measures is proposed in this paper: The use of SVM instead of a single weak classifier AdaBoost algorithm (SVM-AdaBoost) to enhance the classification ability of the classifier. The experimental results shows that the proposed method can achieve better performance than the traditional method in training time and recognition performance.In this paper, the measurement principle of small aperture imaging using monocular camera is used in the vehicle distance measurement. According to the calibration of the camera's internal parameters and the distance measurement model, the distance between the front vehicle and the following vehicle was measured. The experiment of measuring the distance between the vehicle and the front vehicle was conducted by using the video recorder.