Research On Parking Slot Detection And Location Of Automatic Parking Based On Convolutional Neural Network

In the automatic parking system,there are many key problems to be solved,one of which is how to detect and locate the empty parking slot with the marking line around the vehicle quickly and accurately.Traditional vision-based parking slot detection methods have the disadvantages of low detection accuracy,fixed detection scenes,high environmental requirements,and poor generalization capabilities.In order to solve the above problems,a convolutional neural network is used to detect and locate parking slot in this parper.First of all,the Zhang Zhengyou's calibration method was used to calibrate the fisheye camera,and the correction of fisheye image distortion was implemented by a fisheye calibration-based distortion correction model.The experimental results show that the calibration parameters are accurate and the distortion correction effect is good.On this basis,the conversion relationship between pixel coordinate system and vehicle coordinate system was studied,and the plane ranging and measurement of monocular camera based on coordinate conversion were realized.The distance error is within 5cm,which meets the requirements of automatic parking system.For the task of empty parking slot detection,a deep convolution neural network suitable for empty parking slot detection was designed.An empty parking slot data set was produced,which consisted of 16000 RGB color fisheye images,including three categories: empty parking slot,T-shaped parking slot corner and L-shaped parking slot corner,which were manually labeled.The empty parking slot detection model was trained on the parking slot data set,and the detection experiments were carried out in the test set.The experimental results show that the precision of empty parking slot detection is 98.7%,and the missing rate of detection is 1.5%,the detection speed of a single picture is 18 ms,which means that the model has a good detection effect.In the task of locating the center point of the parking slot corner,a small convolution neural network was designed based on the idea of regression of single pixel.The center point data set of parking slot corner was made by 6000 pictures in total,and the center point of parking slot corner were carefully labeled.The detection model of the center point of the parking slot corner was trained on the data set,and the detection experiment were carried out.The experimental results show that log-average missing rate of the detection of the center point of the parking slot corner is 0.40%,the detection speed of a single picture is 18 ms,and the location error is within 5cm.Compared with the traditional detection methods,the parking angle center point detection model has higher detection and localization accuracy with faster detection speed.Finally,the correction of fisheye image distortion,the plane ranging based on coordinate transformation,the detection of empty parking slot and the detection of center point of parking slot corner were integrated.Then,the algorithm of detection and location of empty parking slot were designed,and experiments were carried out on vehicles.The experimental results verify the correctness and feasibility of the algorithm,which shows that the algorithm can detect and locate the empty parking slot in real time and under different ground conditions,of which has a strong generalization ability.