Research On Detecting And Ranging System Of The Leading Vehicle Based On The Sensing Platform Of Machine Vision

With the development of electronic technology, control technology and sensor technology, active safety technology has gradually become an important research direction of the development of automotive industry. This kind of technology can prevent the occurrence of traffic accidents, which is called Advanced Driver Assistance Systems or ADAS for short. Sensors are the basis of ADAS, which mainly include cameras, radars, angle sensors, vehicle speed sensors and so on. These sensors can obtain the internal and external information of the vehicle. Compared with other sensors, the visual sensor can obtain more abundant information, so the visual sensor is more and more widely used in ADAS. This paper establishes the sensing platform of machine vision through using six cameras as visual sensors, and develops detecting and ranging functions of the leading vehicle in this sensing platform. The main work of this paper is shown below.(1)Establish the sensing platform of machine vision. In order to implement two driver assistance functions of visual expansion and machine vision recognition of driving conditions, the sensing platform of machine vision is established in this paper.Six cameras are used to constitute the sensing platform in this system to collect the front, side and rear road information of the vehicle. This paper uses the matching method of camera parameters based on model calculation and vision simulation in CATIA to establish the sensing platform. Firstly, the minimum focal length of the forward-view camera is obtained by model calculation to ensure that the maximal recognition distance from the leading car is greater than the safe distance on the limit road condition. Then angle and disposition parameters of rear-view cameras and side-view cameras are obtained through vision simulation in CAITA, so that the view of rear-view cameras can cover the view field of rearview mirrors ruled in national standard, and there is no blind spot in the view field constituted by the view of rear-view and side-view cameras and the front view of the driver.(2)Develop detecting function of the leading vehicle. The machine learning method based on Haar-like feature and Adaboost algorithm is used to detect leading cars, and the tracking method based on Kalman filter is used to track leading cars. Firstly, this paper trains the classifier through calling opencv traincascade.exe of OpenCV in Command Line Program. Then it implements detecting function of the leading vehicle through calling OpenCV and using C++ language to program on the development platform of VS2010.(3)Develop ranging function of the leading vehicle. This paper selects the ranging method of monocular vision based on geometric model to measure the leading vehicle distance. Firstly, intrinsic parameters of the forward-view camera can be obtained by using calibration method based on OpenCV. Then the pixel coordinate can be transformed into the road plane coordinate through combining the transformational relation deduced by geometric model and intrinsic parameters. Finally, the leading vehicle distance is calculated by putting the road plane coordinate into the distance calculation formula. And the whole development of ranging function in this paper is implemented by using C++ language to program on the development platform of VS2010.(4) Check the view of rear-view and side-view cameras. Through checking the view of rear-view and side-view cameras in the real car, this paper verifies the feasibility of the matching method of camera parameters proposed in this paper. The result of vision check shows that the rear-view image system can replace rearview mirrors and there is no blind spot in the view field constituted by the view provided by the system and the front view of the driver.(5)Conduct the real vehicle test. Through conducting detecting and ranging testes of the leading vehicle on the urban and suburban roads, this paper verifies if detecting and ranging functions of the leading vehicle developed in this paper could meet the application requirement of structured road. The result of test shows that the system can detect leading cars accurately and in real time, and the relative error of ranging results is small.