Study On Image Stitching Of Traffic Accident Scene Based On Geometry Correction

Nowadays, traffic accident is one of social effects of pollution all over the world. More than 3,200 people died and enormous wealth was lost in vehicle accident. Therefore, we should pay more attention to this kind of problem and raise the level of traffic accident analysis to meet the person's legal consciousness. The investigation survey and draft of accident scene are the basis of approving the accident process, and also they are the main evidence of judging the responsibility of an accident. Traffic accident mapping system, which applied efficiently, is a necessary instrument to gain the accurate data of accident scene.It is an advanced method to achieve the plan-form of accident scene by geometry correction for obtaining accuracy data for accident reconstruction. The information that a single photo can provide is limited; several photos geometry corrected will be pieced together to show a complete scene of an accident. Through reading plenty of documents including the relevant system for measuring in other countries, computer image processing technique, computer graphics a method of geometry, the paper gives out the image stitching method, establishes image stitching module of accident scene and develops corresponding software. Finally, the validity of software is tested by experiment.The basic theory for image stitching is image geometry transform while the key technology is image matching and pixel evaluation. Image geometry transformation includes translation, rotation and scaling. The method of image matching adopted in this paper is character point matching. The software offers two methods for picking-up matching point: one is manual mode; the other is automatic mode. When the matching point is clear, automatic mode could be used; if the matching point is blurry, manual mode could be used. After all the matching points are selected, a matching line can be confirmed on each stitching image. The matching points on each image are the same points. So, we can unify reference frames into two images. Firstly, the second image is rotated in order to make two matching lines parallel. Because of the camera's focus, the scales of two images may be slightly different, and then the second image is zoomed in order to make these two matching lines have the same length. Lastly, the second image was translated to match the coordinate.The process of determining pixel value includes two steps. The first step is the interpolation of gray scale; the second one is mapping method of gray scale. The paper used bilinear interpolation method for interpolation of gray scale. There are two methods used for mapping of gray scale: one is pixel carry-over; the other is pixel filling. The former method takes a long time for calculation and the pixel of input image may be mapped out of the boundary of output image. The latter method requires 4 points at least to determine the pixel gray value. This method cannot miss any pixel and it does not calculate any pixel out of the boundary. In this case, I decide to adopt pixel-filling method to carry out pixel gray value mapping.The interpolation of gray scale method cannot be used for color image directly. This paper divides the pixel into three primary colors– red, green and blue firstly, then separate the three primary colors by using interpolation method of gray scaleAlso, based on Photo Stitching model, the software was developed by using Visual C++ 6.0. The software is composed of seven modules, which as the following: image R&W and display module, the geometry correction module, parameter I/O module, image transform module, data exchange module, info-display module, measure module.Finally, this paper analyzed the effect of error through experiment. The experiment was arranged according to four factors–the height of camera while shooting, the focus of camera, the resolution of camera and illumination condition of the scene. The result is that the higher the camera is, the more precise measurement will be; the shorter the camera focus is, the more precise measurement will be; the evener the illumination condition of the scene is, the more precise measurement will be, though the stitching precise has no relation with the resolving power of camera, a resolution above 1280×960 is required.Theoretical analysis and experimental results show that the software of image stitching of Traffic Accident Scene, which based on Geometry Correction, can be used to measure accident scene fast and accurately.