Research And Application Of Measurement System For Vehicle Overall Dimensions Based On Computer Vision Techniques

With the development of society and economy and the rapid increase in vehicle holdings, people enjoy the great convenience and profit brought by transportation development. At the same time, there have been various kinds of illegal actions such as refit vehicles in order to increase the volume. Vehicle overrun and overload have become the major causes of malignant traffic accidents, which have aroused widespread concerns in the community. Vehicle overall dimension measurement is an important way to identify illegal vehicle modification, and it is also an important project in handling vehicle registration and annual inspection. The current commonly used manual measurement such as tapes, rulers and other tools takes high labor intensity and long measuring time, and the results are very likely to be affected by human factors. In order to meet the demand of vehicle management department for automation measurement of vehicle overall dimensions, this paper studies and realizes a measurement system for vehicle overall dimensions based on computer vision, which realizes a non-contact, speedy and accurate measurement of the vehicle overall dimensions. The main contents of this paper are as follows:1. Studyings the algorithm of computer vision in measurement of vehicle overall dimensions from camera imaging model, camera calibration and image rectification. It introduces the image coordinate system, the camera coordinate system, the world coordinate system and their transformational relations, and offers a definition of internal and external parameters of camera on a combination of linear and non-linear models of the camera, describes the camera calibration method based on2D planar target, and elaborates the estimation process of homography matrix, and describes how to correct distortion image to non-distortion image through camera calibration parameters.2. Proposing a measuring method of vehicle overall length. First, we use a single Gaussian model with an anti-shaded background deduction algorithm for image region segmentation, remove image background, and extract vehicle body image, then related methods of vehicle body image stitching are applied: we use Canny edge feature to complete image matching, figure out the translation between images, and complete the image stitching of vehicle body sides, then we complete the stereo matching between images through Harris corner feature, obtain the parallax between two images, and calculate the distance between the vehicle and the camera according to binocular measurement principle, at last, the method of the physical length of vehicle body is presented to obtain the actual measurement results and analyzes them.3. Presenting a measuring method of vehicle overall width. First, we use a simple method of different values between adjacent pixels to extract the edge point of the vehicle top view, use RANSAC algorithm for matching edge points to obtain the linear equation of the edge of the vehicle top view, then we calculate the pixel width of vehicle roof. At last, the actual measurement results of vehicle overall width is analyzed.4. Presenting a complete measurement scheme based on the foregoing methods in the practical measurement of vehicle overall dimensions. It describes systematic design and implementation procedures from the hardware environment erection and software development, and provides the process flow diagram of systematic measurement and processing flow chart of software algorithm.The work described in this paper is a typical application in the combination of computer vision based measuring technology and actual production. Some research results obtained in this paper demonstrate that an effective technology for automated measurement of vehicle overall dimensions can be obtained.