Automatic Parking Slot Detection And Recognition Based On Around View Monitor(AVM) Systems

With the continuous increase of the number of vehicles in China,the urban road space is becoming increasingly crowded,and the number of available parking space is far from satisfying the rapid increase in the number of cars.The narrow crowded parking space and the complex parking environment make it extremely prone to traffic accidents when the driver is parking.It is easy to cause local traffic jam while bringing unnecessary economic losses to the driver.In order to solve the increasingly prominent problem of parking,automatic parking technology has been deeply studied.From the initial Intelligent Parking Assist(IPA)system to fully automatic parking system,automatic parking technology has become one of the important research directions in the intelligent transportation field.At present,the most widely application is an ultrasonic radar-based automatic parking system,which uses the principle of ultrasonic reflection to detect and recognize available parking space to complete automatic parking.However,due to the limitation of the ultrasonic radar working principle,the system cannot accurately detect and recognize free parking space when there is no referable vehicle around the target parking space.The vision-based automatic parking system can effectively solve above problems.This system uses computer vision technology to recognize the ground parking marking line to complete the detection and recognition of available parking spaces.Therefore,this thesis proposes a parking space detection and recognition method based on the Around View Monitor(AVM)system,which can more accurately,safely and quickly realize the detection and recognition of available parking space in complex scenes,and provides powerful support for the subsequent automatic parking system.First of all,in order to obtain more information about the surrounding environment of the vehicle and the characteristics of parking marking lines,in this thesis the AVM system is used to generate a 360° panoramic image around the vehicle,and on this basis,the detection and recognition of parking spaces is completed.In the parking space detection stage,image preprocessing operations are performed on the original panoramic image,mainly including image graying,image denoising and image margining.The purpose is to reduce the amount of data and processing time at the later image processing stage.Then,the line segment detection algorithm based on Line Segment Detector(LSD)is used to detect the parking mark line on the panoramic image map.Compared with the classical Hough transform based line detection algorithm,LSD algorithm,which uses the gradient information to complete the straight line segment detection,has a higher detection accuracy and less computation time.Due to the adaptive error control mechanism of the algorithm itself,it can effectively reduce the occurrence of leak detection and false detection in the parking space detection process.In addition,on the basis of the above-mentioned parking space detection method,this thesis completes the recognition of the available empty parking space.Whether the candidate is an empty parking space is determined by calculating the gray scale change between the length of the parking marking lines and the marking lines.In the existing parking space recognition method,only parked vehicles in a parking space are used as criteria for whether or not the parking space is occupied,and the influence of small obstacles within the parking space(such as locks for parking spaces,traffic cones,etc.)on recognition of empty parking spaces is ignored.This thesis uses image segmentation and binocular stereo vision algorithm to calculate the height of small obstacles to determine whether the target parking space is occupied.In the actual measurement phase,the Microsoft Visual Studio 2013 development kit and OpenCV visual database are used to complete the detection and recognition of parking space.The actual test results showed that the parking space recognition method using small height obstacles can effectively improve the recognition accuracy.