【作者】 李海龙；

【导师】 许洪国；

【作者基本信息】 吉林大学 ， 载运工具运用工程， 2015， 硕士

【摘要】 近年来，伴随着我国经济社会的高速发展，汽车保有量也越来越大。在汽车带给人类极大便利的同时，其所产生的交通安全隐患也越来越多。交通安全问题不仅严重危及生命以及造成大量财产损失，而且在一定程度上也影响着社会的和谐、稳定与发展。针对发生交通事故后的现场勘查方式，现有测量手段，由于主要以人工勘测为主，导致时间浪费过于严重，影响了事故处理进程。随着计算机视觉技术、图形处理技术等学科的发展，将数字摄影测量利用到现场事故勘测已经越来越普遍。相比于手工绘图，数字摄影测量有效地减少了事故勘测的时间延迟，并为绘制事故现场图提供了极大便利。随着普通摄影装置使用量的加大，普通摄影测量的限制也越来越受到事故勘察者的关注。事故现场图像校正一般采用手动选取图像坐标的方式，容易引起选取坐标偏离实际坐标误差较大，影响事故现场的勘测精度，也相应延长了事故处置的时间；同时考虑到摄像机成像系统、取景位置以及镜头畸变等非线性畸变因素的影响，进一步导致精度误差较大。基于这些问题的考虑，本文重点研究关于图像坐标的自动检测手段。采用两种自动检测方法对目标区域的标志物图像坐标进行自动选取操作，保证事故现场勘测精度，而且节约了事故勘测人员的图像绘制时间。图像处理后期利用改进的Tsai两步法进行相机标定，考虑相机径向畸变因素的影响，对相机内外参数进行优化分析，验证Tsai两步法的有效性。为了保证后期自动检测能够顺利进行及其精度，对所拍摄目标图像进行预处理操作，重点从灰度变换、平滑去噪及阈值分割三方面对图像进行处理，保留原始图像特征信息，消除图像噪声等因素影响，并使得图像标志物与背景信息明显区分，为后期自动检测的实现提供理论基础。针对图像自动检测的研究，本文重点采用两种方法，分别是投影自动检测法和基于模板匹配的活动轮廓法。投影自动检测法，首先利用投影法对目标区域进行初定位，获得标志物所在的大致范围，然后通过寻找最大流通域法对标志物进行精确定位，并最终获得标志物图像坐标；通过最终数据验证分析，该方法可靠有效，且精度较高。基于模板匹配的活动轮廓法，首先采用基于灰度的模板匹配法对目标区域进行粗略定位，后期通过基于CHAN-VESE方法来实现标志物的精确坐标定位。在自动检测获得标志物图像坐标后，利用图像二维几何校正模型实现目标区域的俯视图校正。以线性成像模型为基础，采用四参考点法得到空间坐标，建立校正模型，并获得未知参数为后期验证测距提供铺垫。考虑到摄像机镜头非线性畸变因素的影响，导致图像实际坐标与理论坐标之间出现偏差，采用Tsai两步法进行相机标定；利用棋盘模板对相机内外参数进行非线性标定，优化内外参数。为验证自动检测方法的有效性与精度，进行了场地实验；分别从拍摄相机距离地面高度、相机与目标区域的水平距离以及标志物相互间距大小三方面对自动检测精度进行分析；同时验证参考线的测距精度是否和所在标志物覆盖区域的位置有影响进行数据分析；最后，通过与手动选取图像坐标进行对比，验证自动检测方法的有效性。更多还原

【Abstract】 In recent years, with the great development of our country’s socio-economy, the usage of automobile is bigger and bigger. While automobile brings great convenience to human, it gives rise to more and more safety problems. Traffic problems not only seriously endanger people’s lives as well as causing great damage of property, but to a certain extent, affect the social harmony, stability and development. In view of the exploration way after the accident scene, the existing measurement is mainly manual reconnaissance, which leads to wasting time too much and influences the process of the accident. Accompanied by subjects such as computer vision technology and image processing technology, digital photogrammetry has been commonly used to accident survey. Compared with manual drawing, digital photography equipment effectively reduces time delay of the exploration and provides great convenience for drawing the accident scene.Along with the increasing usage of ordinary photography device, the limits of traditional method has been causing more attention from accident tester. Selecting image coordination by manual easily causes bigger coordination error, affects the survey precision, and time loss of accident analysis accordingly. What’s more, the camera imaging system, point position and nonlinear distortion factors such as lens distortion, further lead to bigger correction error of vertical view.Based on the above these factors, the essay focuses on two automatic detection means of the image coordinates, which are used to automatically detect the marker image coordinates of the target area, so as to ensure accuracy of survey and save the time of accident image drawing. Later considering the camera radial distortion factors, use improved Tsai two-step camera calibration to optimize the internal and external camera parameters analysis.In order to ensure accuracy of automatic detection, preprocessing operations are taken, mainly including gray level transformation, noise removal and threshold segmentation, so as to distinguish markers from background clearly, and provide theoretical basis for the realization of the automatic detection.This paper mainly introduces two methods of image automatic detection: automatic test based on the projection and the active contour method based on CHAN-VESE. For the first approach, use the method of projection to rough localization of target area at first, and then make accurate positioning of markers by the biggest circulation domain method, and eventually obtain the accurate image coordinates of marker. For the second method, apply template matching method based on gray level to localize target area roughly, and then achieve the precise positioning of markers based on CHAN-VESE method.Two-dimensional geometric correction mode is used for bird-view correction of the interested area. The space coordinates are firstly gotten according to four reference point method, and correction model is set up. And then attaining unknown parameters make preparation for after-test. Considering the camera nonlinear distortion factors, it will have deviation between actual coordinates and the theoretical image coordinates. The checkerboard template is used for nonlinear calibration of internal and external parameters.In order to verify the validity and precision of automatic detection method, make field experiments to detect precision respectively from the height of camera shooting, the horizontal distance of the camera and mutual distance between target area, and make data analysis on whether the ranging precision of reference line is related to the location of the coverage area. Finally, compared with manual selecting image coordinates, the automatic detecting method is verified to be effective.更多还原