The Study Of Visual Detection In The Automatic Paraller Parking Process

To detect the position and orientation which vehicle opposite to the parking spacesand the obstacle position is an important component of automatic parking environmentperception systems.To get these detection datas what belong to the context-awaredata,which absolutely can provide some imformations for the vehicle obstacle avoidanceand the parking path planning.This paper is based on machine vision technologyresearches on the visual detection methods of the position and orientation detection thatvehicle opposite to the parking spaces and the obstacle position detection.The maincontents are as follows.1) In order to acquire images in the open air,we put up an experimental car that canacquire images through wifi.The experimental car can work3-4hours in the open air.wepropose a new parallel parking path which is based on the two existing parallel parkingpaths.This parking path divides the whole parallel parking path into five stages.Eachstage corresponding to different region of interest are located in a specific rectangularregion of the image.The experimental results show that this method can effectivelyextract the region of interest which contains the parking spaces.2)Firstly,we take some pretreatments for the region of interest that contain theparking spaces.The pretreatment includes graying,median filtering and thresholdsementation.After that we can get a binary image of the region of interest which containsthe parking spaces.Secondly,we can separate the parking outside line and the bottom linebecause the two parking lines show different slope and intercept in the images.Finally,weextracte the feature points of each parking line from the images which only contains oneline.The experimental results show that this method can effectively separate the twoparking lines and extracte the feature points of each parking line.3)Firstly,we put some scene points which can be served as a calibration target on theground,calibrate the camera with BP neural network,thus we can establish a BP neuralnetwork model that reflects the relationships between the image pixel coordinates and theworld coordinates.Secondly,we get the world coordinates of the feature line whichextrated in the second step,then using the method of the least squares to fit the parkinglines in the world coordinate system.Finally,we estimate the position and orientation thatthe vehicle opposite to the parking space. The experimental results show that this methodcan effectively estimate the position and orientation that the vehicle opposite to theparking space.4)Firstly,we extract the line of intersection what belong to the ground and the obstacles,then getting the endpoints of the line of intesection.Secondly,we can use thecalibration camera medel to get the world coordinates of the two endpointswidth.Finally,we obtain the obstacle position and width with the derivation of the formula.The experimental results show that this method can effectively obtain the obstacleposition and width.