Research On Multi-sensor Combined Automatic Parking Perception And Path Planning

As the number of motor vehicles increases,more and more drivers are troubled by the difficulty of parking.The automatic parking function in intelligent vehicles has gradually attracted the attention of drivers.At present,the algorithm that uses cameras to identify parking spaces can detect parking lines under the condition of simple environmental information and moderate brightness.If there are other complex shapes or objects close to the color of the parking space line near the parking location,it is likely that the parking space will not be detected.In the existing researches on automatic parking path planning,most scholars do not pay attention to the problem of pivot steering when planning the paths.In order to solve the above problems,this paper designs a parking space detection method combining deep learning and Open CV,which enhances the ability of automatic parking sensing module to resist environmental interference.At the same time the distance coordinate position of parking corner can be found more accurately.The problem of pivot steering in automatic parking paths planning is solved by means of easement curve and polynomial scattering points method.The main research contents are as follows:(1)Collect panoramic top view images.The original images are captured by four fisheye cameras installed on the car.The distortion of the fisheye cameras is corrected by using a polynomial model and bilinear interpolation.I use the grayscale world method to balance the brightness of the four images.The squint image is transformed into a top view image by solving the perspective transformation matrix.A panoramic top view image is obtained by using equal proportion splicing and weighted fusion.In top view image,I convert pixel coordinates to distance coordinates on actual plane.(2)Recognise parallel parking space and vertical parking space.The deep learning model is trained to recognise the approximate location of parking space.The parking space images are taken by the experimental car,the data set of the paper,mobile phone camera.Parking space image is processed by grayscale and image filtering.After parking space background is extracted by morphological operation,the background will be removed.The parking lines are achieved after binarization treatment.I use polynomial to fit the vertical and horizontal parking lines and find the corner points of parking lines.Then the pixel coordinate of the parking space image is converted into the distance coordinate of the relative car in the panoramic top view image.(3)Utrasonic radar detects barrier.Based on Freescale micro control unit development board,external ultrasonic radar and temperature sensor are installed.A temperature sensor collects temperature and compensates for the speed of sound.The ultrasonic radar obtains the distance of the obstacle through the duration of the high level and the compensated sound velocity.I use a serial port to transmit temperature and distance information.(4)Sensors combination to detect parking spaces.The fisheye cameras detect the parking lines and find the corners coordinates of the parking space,while the ultrasonic radar detects whether there is an obstacle in the parking space.Ultrasonic radar detects the vertical distance between obstacle to the side of the vehicle.Ultrasonic radar and fisheye cameras are combined to detect the distance between obstacle and the rear edge of the parking space.(5)Design automatic parking path.The distance coordinates of parking corners are input into the path planning module.Appropriate corner is selected as the origin point of coordinate system.Based on the vehicle low-speed kinematics model,the parking paths of vertical and parallel parking space are designed in MATLAB by means of easement curves and polynomial scattered points.