Design Of Parking Spaces Visual Detection And Positioning System Of Automatic Parking Based On Deep Learning And OpenCV

The rapid growth of motor vehicle ownership has brought tremendous pressure to the traffic environment,at the same time,the difficulty of parking is aggravated.As the most anticipated but immature assisted driving function for consumers,the automatic parking system is of great significance to solve the problem of parking difficulty.The automatic parking system is mainly divided into two categories,the first one is based on ultrasonic,and the other is based on images.The ultrasonic-based automatic parking system can not detect parking spaces without reference,there are strict requirements on the position and angle of vehicles placed on both sides of the parking space,it can not make judgments on the oblique parking spaces,and it is prone to cause misidentification.The vision-based automatic parking system provides the possibility to solve the above problems.At present,most of the vision-based parking space detection algorithms are effective only under the conditions of good lighting conditions and less shadows and simple information contained in the images.Some studies only detect the sidelines of the parking spaces,and do not design how to judge the specific parking spaces and complete the four vertices positioning of the parking space according to the detected sidelines,because it is a complex logic problem,especially when the image contains a large number of parking spaces and the parking space distribution is irregular.This paper designed and implemented a parking detection and positioning system based on deep learning and OpenCV,and this method can effectively solve the problem that the parking spaces cannot be detected and positioned under the condition that the input image contains complex backgrounds.The main research contents are as follows:(1)In order to detect and position the parking spaces around the vehicle as many as possible and as early as possible,a 360-degree panoramic view system with a wide field of vision was designed and built for image acquisition.Firstly,four fisheye cameras were used to capture the images of the surroundings around the test vehicle.The cameras parameters were obtained by using Zhang Zhengyou’s calibration method,the distortion correction was performed on the pictures according to the parameters obtained by the calibration.And then,the top view transformation algorithm was usedto perform the top view transformation on the corrected image to obtain a top view perpendicular to the ground.Then a coordinate system for mosaic was designed,the overlapping areas of the images were fused by weighted average fusion algorithm.Finally,a 360-degree panoramic view was obtained,and the panoramic view images were used as the input of the parking spaces detection system.(2)In order to solve the problem that the parking spaces cannot be detected and located when there is a large amount of complex information in the image,this paper designed and implemented a parking space detection system based on deep learning.Firstly,two models for parking spaces detection were built by using deep learning method.The designed 360-degree panoramic view system was used for image acquisition to make training samples and testing samples.Then,the prepared samples were used to train the built test models.Finally,the performance of the trained model was tested and validated with typical pictures,and the results were analyzed in depth.(3)After using the trained deep learning model to complete the parking spaces detection,many small areas in the image were extracted,and each of them contains one independent parking space.Based on this,this paper designed a set of parking space positioning algorithm based on OpenCV to locate the four vertices of the parking space in each extracted small area.Firstly,the image was processed by using graying algorithm and filtering algorithm to reduce the image noise;For the image with uneven illumination or a large number of shadows or complex background information,the common binarization method can not get the complete parking space from the original image.This paper used opening operation to get the background of the input image and removed that from the original image,after that,the binarization operation was performed;Then,a parking space extraction method based on connected regions was proposed according to the characteristics of the parking space;Finally,the positioning of the the parking space’s vertices were completed by Hough transformation,and the four vertex coordinates of the parking space were finally output.Four coordinate values of the parking space in the panoramic image were obtained by coordinate transformation.The experimental results show that the panoramic view system designed in thispaper can achieve a good result of images stitching and it can be used as the input of the parking spaces detection and positioning system.The parking spaces detection system based on deep learning can obtain high detection accuracy of the parking spaces under the condition that the images contain complex information,and the mean average precision of the parking spaces detection reached 89.30%.The designed parking spaces positioning system can realize the positioning of the parking spaces,and the four vertices of each parking space were positioned accurately,that can be used as coordinate input for the automatic parking system.