Abstract: To improve the single application adaptability of machine vision technology in the field of automatic parking to the orientation detection of the target empty parking space, a new machine vision detection method for the relative position of car body was investigated based on the close-range image of binocular lens. The close-range image of the parking space was collected through the binocular lens. Through the image processing of image correction, image mosaic, inverse perspective transformation and image binarization, the contour edge of the target empty parking space on the top plane and the position point of the binocular lens were extracted. The coordinate system of the vehicle body relative to the target empty parking space was established, and the coordinates of the binocular lens in the two-dimensional coordinate system of the parking space were determined through visual recognition and pixel point coordinate calculation. Based on the known parking space and car size, the position of fisheye lens on the car body and other information, the calculation relationship model between the azimuth and distance of the car body relative to the parking space was established by geometric method. The results show that the maximum relative error of azimuth detection is 9.16%, and the maximum relative error of range detection is 3.89%, which improves the adaptability of target empty parking space detection.