Research On Getting Car Position Algorithm Using Machine Vision For Automatic Parking

Automatic Parking System has different kinds of sensors to get the information of vehicle and its surrounding environment, for the parking system making the basis strategy. The use of visual sensor in automatic parking system can get image information that other sensors can’t get. Now the use of visual in automatic parking system is less than other sensors, the research in domestic is also less. Therefore, do the research on the machine vision for getting car position algorithm in this paper.The method getting space location with machine vision, according to the number of cameras can be divided into monocular vision, binocular vision and more eye vision. Because monocular vision computing space position need exact camera movement parameters, and camera’s accurate motion is relatively difficult to control, binocular vision and more eye vision use fixed camera position, so this paper use binocular vision method for car position caculation.To calculate space position with binocular vision, first to acquire binocular cameras’ parameters. Camera calibration algorithm can be divided into traditional calibration algorithm, the self-calibration algorithm, the active vision based calibration algorithm, etc. The traditional calibration algorithm need target, calibration process is complex, calibration precision is higher; Self-calibration algorithm does not need target, is convenient and flexible, but lack precision, stability and robustness is bad; Active vision based calibration algorithm is based on the active control of camera movement, robustness is good. Because in this paper camera position and the focal length is fixed, an offline calibration is enough to use, choose to use traditional calibration algorithm for camera parameters. Traditional calibration algorithm have TSAI calibration algorithm, ZhangZhengYou calibration algorithm and so on. In order to get the adaptive camera parameters for this paper, this paper use the above two methods for calibration, through the experiments verify the accuracy of the ZhangZhengYou calibration algorithm is low, and TSAI calibration algorithm need to know the optic center position, its accuracy directly influence the calibration precision. In order to get the optic center position, use another traditional calibration algorithm in the reference document [43], and use its optic center result in TSAI calibration algorithm. Through the experiment the calibration result of this way can meet the accuracy demand in this paper.With camera parameters, to calculate parking space position, it still needs parking feature points in binocular vision images. This paper take parking feature points as four corner points, because the parking has obvious shape characteristic, this paper use shape feature extraction algorithm extract parking characteristics, main steps include edge feature extraction, quadrilateral feature extraction, linear feature extraction. According to the rectangular characteristic of the parking, use line cross to determine corner points. This paper proposes a correlation based statistical algorithm, which can extract more integrated parking line from the quadrilateral edge in the distortion image. The experimental results show that the method can more accurately extract parking feature points.With binocular vision model and parking feature points the parking position can be calculated. Due to the parking has different locations, the parallax in the image is different and3d coordinate calculation accuracy will be also different. This paper analyzes the3d coordinate calculation accuracy of spatial points in different space position. For camera resolution limit, the3d coordinate calculation result has low accuracy when space point is farther (more than4m). In order to improve the3d coordinate calculation precision of space points, this paper proposes a method combined the inherent rectangular characteristic of parking with genetic optimization algorithm. To varify parking corners’3d results whether meet the rectangular feature constraint or not, if not, use the genetic algorithm to search the best parking position in its nearby space.On the basis of the study, make a series of ground point and parking experiment, the experimental results show that, using the method in this paper can get the car position relative to the parking information accurately with binocular vision when the distance between car center and parking center is in the range of5-10m, and the computation precision is convergent as the car getting close, which can meet the requirement of getting car position in the automatic parking system.