【作者】 李立冬；

【导师】 刘教民；

【作者基本信息】 河北工业大学 ， 计算机应用技术， 2005， 硕士

【摘要】 原有的基于单幅图像确定目标方位的交通事故处理系统采用普通数码相机拍摄现场,用计算机图像处理技术和摄影测量原理来分析交通事故现场,计算所需要的各种数据、各个方位的尺寸,包括地面上任意两间距离,点到直线的距离,最终自动生成事故现场图。该系统确实提高了交通事故的处理效率,有效减少了交警在交通事故处理中的劳动强度和人为因素,但对恶劣天气下所拍照片的处理效果不是很好,测量精度有待进一步提高,各种数据的测量也仅限于地面。 因此,本文从以下三个方面入手,提高原有系统的性能。首先,由于测量的准确性很大程度上取决于相机标定,故采用自适应阈值图像分割技术实现图像的二值化,进而用Prewitt算子针对标盘局部区域进行边缘检测,用霍夫变换检测出标盘上的直线,以检测出的直线来标定相机。把这些图像处理过程局限于对比度较高的标盘区域,不仅可以实现有效的噪声消除和边缘检测,而且可以大大提高霍夫变换检测直线的效率。其次,以三维标定物代替平面标定物,增加了标定点个数,用最小二乘法求解相机参数,提高了准确度;同时在成像模型中考虑非线性畸变,使成像模型更符合实际情况,提高事故现场的测量精度。最后,在平面测距基础上,增加一个约束条件,求解线性方程组即可得到任意空间点的坐标,为事故现场的三维重建奠定了基础。更多还原

【Abstract】 The original Distance Measure System of Traffic Accident Based on Single View is a system that employs common digital camera taking pictures on the spot, calculates useful data including the arbitrary distance from one point to another, from a point to line and so on by using the image processing technology and photogrammetry, and then produces the responding traffic accident graph automatically. Finally the analysis can be carried out according to them. During its using in true accidents, its great performances are embodied in improving efficiency of dealing with traffic accidents, reducing labor intensity and human-factor influences. But at the same time some problems are also exposed, such as the processing results under bad weather conditions are not satisfactory. And its measurement accuracy needs to be improved. Furthermore, all measurements are limited in a plane referred as the road flat in the system.So in this article three aspects are adapted to improve the performances of the original system. Firstly, the accuracy of the measure system greatly depends on the camera calibration, so self-adaptive method to determine the threshold is adapted in image segmentation. Then the Prewitt operator is used to detect the edge within the local area of the calibration object. The lines are detected through Hough transform to calibrate the camera. All processes are limited in the high contrast area of the calibration object, which helps to clear noises, detect edge effectively and enhance Hough transform greatly. Secondly, three-dimensional object instead of two-dimensional one used to calibrate can also get better accuracy, for there are more reference points available and least square approximation is used. On the other hand, nonlinear distortion is taken into account to make the camera model accord with the fact better. At last, an additional restriction is set and then any point’s coordinate can be calculated through solving linear equations, which can be the foundation of the three-dimensional reconstruction.更多还原