Research On Application Of Camera Self-calibration Technology In Photogrammetry Of Traffic Accident Scene

In recent years, with the enhancing of people's jural consciousness, the traffic cases that need analyzing scientifically have been increasing years by years in case of traffic accident. Requests on the disposal procedure in traffic accident have been becoming higher and higher. Presently, The method of manual-measure cooperated scene photography is mostly adopted in the scene reconnaissance of traffic accident. This method has limitation whether in efficiency and quality of scene mapping or reducing delay loss etc, which particularly doesn't fit with the scene mapping on the freeway where traffic volume is great and speed is high. With the improvement of electronic computer and image treatment technology, photogrammetry has been a good method of solving the above problems. Two-dimension or three-dimension reconstruction of traffic accident scene is made based on the photographs of accident scene which has signification meanings on dealing with traffic accident correctly and timely. Speediness and agility camera self-calibration just can meet are required in traffic accident scene photogrammetry. In order to achieve the aim that is dealing with traffic accident quickly to resume the normal traffic situation of the road, linear least-squares of tradition camera self-calibration and RAC methods to calibrate camera interior parameters indoor are adopted to resolve the problem owing to much unknown parameter in self-calibration method which can improve definition of self-calibration method. Then camera exterior parameters are calibrated with the self-calibration method of fundamental matrix. According to the need of 3D-reconstruction in the traffic accident scene, This dissertation is consisted of the following chapters interrelated to each other: (1) Camera imaging model. Photogrammetry of traffic accident scene is based on the camera imaging model. Based on the linear camera imaging model, the imaging process is detailedly analyzed, relations among the world coordinate system in course of imaging process, camera coordinate system and image coordinate system are studied and the theory of space distance 3D-reconstruction is analyzed. Because of the lens distortion and install errors, there are some difference between imaging with linear model and fact projection. Thus the non-linear camera models are given in this chapter. (2) 3D-reconstruction theory based on photography image of fundamental matrix. In the two uncalibrated images for the same aims, epipolar constraint is only information that can be known. It is fundamental matrix that describes this relationship. Essential matrix is calculated after the fundamental matrix gotten. By singular value decomposition to essential matrix, exterior camera parameters are gained. Then based on the linear camera model, intrinsic camera parameters are gotten with linear least-squares and RAC calibration technique. Consequently get the transformation matrix of projection form the two photograph images can be gotten. (3) The estimation method of the fundamental matrix. Homography matrix method, eight analysis method, weighted normalization method and least median of squares (LMedS) are studied in this chapter, based on which, carried through precision tests of fundamental matrix indoor with the different cameras are carried out. According to the result of tests, could get much the error of fundamental matrix is higher with homography matrix method and eight analysis method, enhances distinctly the estimation precision of matrix is remarkably improved with weighted normalization method, precision with east median of squares method, and precision is correspondingly higher with the high resolution camera. (4) The impacts on precision of 3D reconstruction of traffic accident photograph images with the different fundamental matrixes method,different intrinsic camera parameters estimation, different choice of calibration dots. To realize the manual choosing calibration dots, the calibration block is designed. For the plane is required on photography scene in order to estimate fundamental matrix with homography matrix method, the plane is designed. According to the tests, whether manual or automatic choosing calibration dots, whether using the linear least-squares technique or MAC method to calibrate intrinsic camera parameters of matrix, error of 3D reconstruction space is higher with homography matrix method and eight analysis method and error of 3D reconstruction of space is smaller with the weighted normalization method and stable. When using manual choosing calibration dots to simulate 3D space on the scene of traffic accident, the error is higher with least median of squares technique. When using automatic choosing calibration dots to simulate 3D space on the scene of traffic accident, the error is smaller with least median of squares technique and stable. The errors of 3D reconstruction with the method adopted in this dissertation mainly compose the...