Dissertation On Automatic Parking System Based On The Fusion Of Millimeter Wave Radar And Visual Information

With the improvement of the quality of life of residents,the number of car ownership in China is increasing year by year.The traffic pressure in cities is increasing day by day.Whether on road or in the parking lot,the traffic environment is becoming more and more crowded.Crowded parking spaces bring difficulties and pressures to both novice and old drivers with varying degrees.The ratio of traffic accidents that are caused by parking increases year by year.As the last link of automatic driving technology,automatic parking technology not only has an important impact on passengers' ride experience,but also has an important significance for the development of automatic driving technology.In this paper,an automatic parking system based on millimeter wave radar and visual information fusion is studied.The main research work includes the following:Aiming at the shortcomings of the installation position of the cameras in the market at present,this paper puts forward a new type of camera installation method,then uses the fisheye module of OpenCV3 to calibrate the internal paramaters of the fisheye camera.By making full use of the advantages of the installation position of the lateral camera,the stitching of the images is carried out,and a good stitching effect is obtained.For the detection and recognition of parking space funciton,this paper analyses the difference between parking and lane detection,points out the complexity of parking compared with lane detection and proposes to do the needless lines filter,clustering and parking lines fitting based on the traditional Hough transform line detection algorithm.Through this method,a group of lines which can represent the parking lines are extracted successfully.Then the parking sapce recognition is carried out according to the extracted line segments which including candidate pair selection,candidate pair combination and effective parking space recognition.Then the gradient descent algorithm is used to calibrate the millimeter wave radar and the radar is used to realize the function of idle parking space recognition and obstacle stop avoiding.Secondly,by establishing a detailed parallel parking mathematical model,the constraints of double arc path planning are derived and the minimum parking space length and the accurate mathematical description of the initial parking area are calculated through the constraints.A semi-real-time multi-segment parallel parking path planning method with arbitrary initial position and posture is proposed by using the parkable initial position.For the control algorithm part,firstly,the dead reckoning method is introduced and the experiments proves that its positioning accuracy meets the requirements of automatic parking.Then the control algorithm is introduced from the lateral control and the longitudinal control of the vehicle.For horizontal control,a combination of preview control model and Stanley control model is proposed.For vertical control,the specific realization methods of vertical control are introduced from two aspects: expected speed decision-making and speed tracking.Then the Prescan simulation software is used to validate the path planning and path tracking control algorithm.It is found that the linear path control accuracy is less than one centimeter,and the double arc path control accuracy is about five centimeters.As a whole,it achieves better control effect.Finally,the whole automatic parking system is experimented by building a real vehicle experimental platform.It is found that the effect of real vehicle control is not much different from the simulation results.This phenomenon fully proves the importance of simulation experiment in the development of automatic driving function.