Study On Key Problems Of Three-Dimension Reconstruction Of Traffic Accident Scene Based On Common Camera

Recent years, the numbers of traffic accidents were increased. The dissensionabout traffic accident has been more and more. It is required to improve ability to dealwith traffic accident. Accident reconstruction is the base of dealing with trafficaccident. The accuracy of accident reconstruction depends on the integrity and detailof information about accident. The basic information for a typical accident usuallycomes from two sources: the accident or police report and photographs that may havebeen taken of the vehicles and scene. The police report may or may not have adetailed record of all needed information. When traffic accident was going to bereconstructed, some information could not be found, that may include involvedvehicles positions of rest, skid marks, gouges in pavement, fluid trails, debris,bloodstains, or off-highway ruts. The photographs of accident scene and vehicle havedetail information about these. So it is important for traffic accident reconstruction todetect information from photographs. Because lacking techniques, photographs of accident is often only used toanalysis accident qualitatively. More information can not to be used. Now it can beused to detect three-dimension (3D) information from photographs of accident as thecomputer vision techniques improved. The key problems of 3D reconstruction fromphotographs took by common camera are camera calibration and image matching.The dissertation studied on the basic techniques of 3D reconstruction in order to findproper theory and arithmetic for 3D reconstruction of traffic accident scene. Ａｃｃｏｒｄｉｎｇ　ｔｏ　ｔｈｅ　ｎｅｅｄ　ｏｆ　３Ｄ　ｒｅｃｏｎｓｔｒｕｃｔｉｏｎ　ｏｆ　ｔｈｅ　ｔｒａｆｆｉｃ　ａｃｃｉｄｅｎｔ　ｓｃｅｎｅ　ｕｓｉｎｇ　ｃｏｍｍｏｎ　ｃａｍｅｒａ,　ｔｈｅ　ｓｔｕｄｉｅｓ　ｉｎｃｌｕｄｅｄ:　 (1) 3D reconstruction of photographs is based on the camera models. Thesimplest camera model is linear transformation without modeling the lens distortion.According to the linear camera model, the dissertation analyzed the geometric relationbetween a 3D object point and its 2D image projection, and discussed the relationsamong the world coordinate system, camera coordinate system and image coordinatesystem. Because of the lens distortion and install errors, there are some differentbetween linear model and fact projection. The non-linear models were discussed inorder to improve accuracy of reconstruction. Based on linear camera model, thetheorys of 3D reconstruction of point, 3D reconstruction of line and 3D reconstructionof quadratic curve were discussed. Because of 3D reconstruction of point is the base 139吉林大学博士学位论文of 3D scene reconstruction, linear least-squares technique, eigenvalue analysismethod, iterative linear method and non-linear method of it were discussed. (2) Edge is important feature in image, and it includes useful information. Intraffic accident image, many features detection must base on the edge detection. Edgedetection includes step edge detection and ridge edge detection. Both step edgedetection and ridge edge detection were discussed. The Infinite SymmetricExponential Filter (ISEF) is fast order optimality for step edge detection, and it can berealized recursively, and its parameters can be adjusted easily. The ISEF was used todetect step edge. After filtered by Infinite Impulse Response (IIR) filter, ridge edgewas detected by using non-maximum suppression and threshold. In order to realizearithmetic easily, ridge edge response and orientation calculating methods wereimproved. ISEF was used to detect edge orientation after IIR filter was used. (3) Line is one of the most important edges in image. In traffic accidentphotographs, Line edges include useful information. It is important to detect line fromaccident photographs. The line detection methods were discussed, and a line detectionmethod was presented, which based on chain codes detection. C...