Research On Key Techniques Of 3D Vision Swift Reconnaissance System Of Traffic Accident Scene

In recent years, with the fast development of our country's economy, the vehicle quantities has increased year by year, car has become indispensable vehicle when we travel,but it causes a lot of traffic accidents in the same time. The traditional way, using artificial to deal with the scene of accident, is not only has low accuracy and easy to leak test, but also very slow, so it is far from meeting the demand of society today. How to process the accident scene fast and accurate has became an issue of various countries' relevant researchers and traffic department personnel. Photogrammetry is an intelligent method, which can dispose the accident scene fast and efficient, using CCD camera to get same images in the traffic accidents scene, then use specific software to process the images, after that, the three-dimensional information is get easily. This method is easy to operation, and the process is fast, beside that, the image record of the accident scene information complete and detailed,so it is facilitate to measurement the scene when you want and eliminate errors made by artificial measurement.However, the previous accident scene rapid reconnaissance system of vision are still some defects, so this topic presents a new vision system, which is based single mobile visual mode. Using a CCD camera to image the coded targets, which is distributed in the accident scene beforehand, in several locations and orientations, then using the uniqueness of coded targets can realize the image registration and image mosaic quickly, using the anchor point of coded targets to calculate the camera external orientation, and with the spatial ratio constraint provide by coded targets, in the last, using bundle adjustment algorithm to process these images, and calculate the 3D coordinate of the space feature point in the accident scene.The results can be used for reconstructing 3D information of the accident scene, and it is helpful for analyzing the dynamic occurrence process of accident, so it provide evidence for accident responsibility identification in the last. The experimental results demonstrate that the method is feasible and flexibility.