基于实虚靶标融合及无人机摄影的交通事故现场勘测技术研究

发布时间：2018-01-15 07:08

本文关键词：基于实虚靶标融合及无人机摄影的交通事故现场勘测技术研究　出处：《吉林大学》2017年博士论文　论文类型：学位论文

【摘要】：交通事故现场勘测是道路交通事故处理的首要问题,对事故成因的全面细致分析具有重要意义。基于单幅图像的摄影测量二维直接线性变换(2D direct linear transformation,2D-DLT)方法,能从事故现场拍摄的像片中获得车辆停止位置、制动拖印、碎片散落分布位置等重要信息,受到了交通事故现场勘测领域的关注。但由于在勘测时标定物上的特征点不能分布于交通事故现场关注区域的四周场景,以及无法克服拍摄倾角对勘测精度的影响等原因,导致适用性受限,实用性不强。因此,应对摄影测量2D-DLT方法进行改进,为交通事故现场勘测提供新的技术方法和理论支持。本文以交通事故现场勘测技术为切入点,通过构建实虚靶标融合模型,获得了大范围的虚拟标定物。为提高实虚靶标融合的实施效果,设计了实标定物,研究了其上特征点的检测方法,并进行了实虚靶标融合的布局分析。为提高摄影测量2D-DLT方法的适用性,提出了基于实虚靶标融合的改进2D-DLT法,克服了单一标定物外延区勘测精度差的缺陷。为提高摄影测量2D-DLT方法的实用性,提出了小型无人机(small unmanned aircraft vehicles,SUAV)摄影2D-DLT测量法,有效解决了使用普通相机采用摄影测量2D-DLT方法勘测交通事故现场时,由于拍摄倾角小造成的勘测精度低等缺陷。具体研究工作如下:实虚靶标融合模型的构建。为了获得与交通事故现场勘测相适应的较大范围虚拟标定物,在研究相机成像过程及镜头畸变因素的基础上,通过摄影时布设在交通事故现场中的各个实标定物相对于相机坐标系的旋转矩阵和平移向量,将多个实标定物的位置统一到相同坐标系下,构建了实虚靶标融合模型。从相机内方位参数的计算、实标定物位姿的计算及模型参数的优化三个方面,研究了实虚靶标融合模型参数的计算方法。以摄影时实标定物上自身尺寸的欧式距离不变、实标定物之间满足共面关系及控制点的重投影误差最小为约束条件,优化了实虚靶标融合模型参数。最后,以仿真试验验证了实虚靶标融合模型构建及参数计算的可行性和稳定性。实虚靶标融合的实施及布局分析。在实虚靶标融合模型的构建过程中,实标定物是重要基础。通过分析交通事故现场特点,设计了平面结构的黑白相间棋盘格标定物。使用中值滤波方法对现场图像进行预处理,突出图像中感兴趣的特征后,采用半自动检测(semi automatic detection,SAD)方法提取实标定物上的特征点。通过实标定物上特征点提取试验,与Harris检测方法对比,可有效抑制非精确控制点的出现,提高实标定物上特征点提取的准确性。为提高实虚靶标融合的实施效果,通过室内试验对实虚靶标融合的布局进行分析。研究结果表明,对于现场固定区域布设4个分散的实标定物,能以最少的实标定物数获得最大的虚拟标定物覆盖面积。基于实虚靶标融合勘测事故现场。将实虚靶标融合应用于交通事故现场勘测,克服单一标定物外延区勘测精度差的缺陷,提高其适用性。通过研究摄影测量2D-DLT方法原理、非线性优化方法以及传统的控制点坐标获取方法,建立了实虚靶标融合后控制点的优选策略,从而获得事故现场关注区域四周的控制点坐标数据,提出了基于实虚靶标融合的改进2D-DLT法。通过现场试验分别考查了实虚靶标融合后建立的虚拟标定物所占区域大小、虚拟标定物所占区域大小相同时实标定物面积大小以及对比使用单一实标定物与实虚靶标融合对勘测结果的影响。试验结果显示:虚拟标定物所占区域越大,待测区域的测量相对误差平均值越小,摄影测量的结果越好;在虚拟标定物所占区域大小不变的前提下,随着实标定物面积的增加,待测区域的测量相对误差平均值变化不大,摄影测量的结果基本不变;采用实虚靶标融合比采用单一实标定物进行摄影测量的结果更好。从而验证了基于实虚靶标融合的改进2D-DLT法的优越性。最后,对一个案例应用结果分析,并证明了改进2D-DLT法的可行性。SUAV摄影勘测事故现场。SUAV摄影可方便地实现从低空俯视拍摄地面图像,而且成像分辨率现已满足有关标准的要求。将SUAV摄影应用于交通事故现场勘测,抑制使用普通相机采用摄影测量2D-DLT方法勘测交通事故现场时,拍摄倾角对勘测精度的不利影响。为减少2D-DLT方法计算中的病态方程求解问题,提出了先将像点的量测坐标值和物点的世界坐标值归一化处理,然后采用奇异值分解的方法计算DLT参数的SUAV摄影2D-DLT测量法,并通过仿真试验证实了该方法的有效性。为了给SUAV摄影2D-DLT测量法的实施提供指导,研究了摄影高度的估算方法。最后,将实虚靶标融合用于SUAV摄影2D-DLT测量法,构建了SUAV摄影勘测事故现场的完整方法,设计了方法流程、硬件设备以及测试程序。通过现场试验验证了SUAV摄影勘测事故现场的有效性,测量数据的平均误差小于1%,比使用普通数码相机采用地面摄影测量的平均误差进一步减小。SUAV摄影勘测事故现场,能有效解决由于拍摄倾角小造成的图像校正精度低、场景覆盖面小及可用信息恢复少等缺陷,提高交通事故现场勘测的精度,增强摄影测量2D-DLT方法的实用性。
[Abstract]:The traffic accident scene survey is the primary problem of road traffic accidents, a comprehensive and detailed analysis on the cause of the accident is of great significance. Photogrammetry 2D direct linear transformation based on single image (2D direct linear transformation, 2D-DLT) method, get the vehicle stop position to take pictures from the scene of the accident, the brake drag India, important pieces of information scattered location, by field survey of traffic accident scene. But due to concerns around the scene calibration feature points on not distributed in the region of interest in the investigation of traffic accident scene, and unable to overcome the influence of shooting angle on surveying precision and other reasons, resulting in limited applicability, practicality is not strong. Therefore, to improve the response of photogrammetry 2D-DLT method, support new methods and theories for traffic accident scene survey. Based on the traffic accident scene survey As a starting point, through the construction of virtual target fusion model, the virtual calibration object in wide range. In order to improve the effect of the implementation of the real and imaginary target fusion, is designed to achieve calibration, the research on feature point detection method, and analyzes the real and imaginary target fusion layout. In order to improve the applicability of photography the measurement method of 2D-DLT, this paper proposes an improved 2D-DLT method based on the fusion of real and virtual target, overcoming the single calibration accuracy of survey material epitaxial region defects. In order to improve the practicality of photogrammetry 2D-DLT method, proposed the small UAV (small unmanned aircraft vehicles, SUAV 2D-DLT) photography measurement method, effective solution to the use of ordinary camera the use of photogrammetry 2D-DLT method to survey the traffic accident site, due to the low precision of surveying defects caused by small angle shot. The specific research work is as follows: the real and imaginary target to construct the fusion model. In order to get The larger range and adapt to the traffic accident scene survey of the virtual calibration object, based on the factors of distortion imaging process and camera lens, through various photography layout at the scene of the traffic accident in the calibration object relative to the camera coordinate system rotation matrix and translation vector, a plurality of position calibration object to the same coordinate. Under the construction of real and virtual target fusion model. From the calculation of orientation parameters in the camera calibration, optimize the pose calculation and model parameters of the three aspects, studies the method of calculating the real and imaginary target fusion model parameters. Real time calibration of Euclidean distance with photography on its dimension unchanged between real calibration things that satisfy the minimum reprojection error between coplanar and control point as the constraint condition, optimization of the real and imaginary target fusion model parameters. Finally, the simulation results show that the real and imaginary target fusion model and The feasibility and stability of calculation parameters. The real and imaginary target fusion implementation and layout analysis. In the process of building virtual target fusion model, real calibration is an important basis. Through the analysis of the traffic accident scene, design of the plane structure of the black and white checkerboard calibration. The preprocessing of the scene image by median filtering the method, characteristics of interest outstanding image after using semi automatic detection (semi automatic detection, SAD) method to extract the real calibration feature points on the calibration test. The feature points on Extraction and Harris detection method of contrast, can effectively inhibit the emergence of accurate control points, improve the calibration accuracy of extraction the feature points on. In order to improve the effect of the real and imaginary target fusion, based on the real and imaginary target fusion layout of the indoor test analysis. The results show that, for fixed site area layout 4 The dispersion of real target, with a minimum number of real calibration object to obtain the maximum virtual calibration object coverage area. The real and imaginary target fusion accident scene investigation based on virtual reality. The target fusion is applied to the traffic accident scene survey, overcome the single calibration accuracy of survey material epitaxial region defects, improve its applicability by 2D-DLT principle. The method of photogrammetry, nonlinear optimization method and the traditional method to obtain the coordinates of control points, to set up an optimum control strategy of point target fusion of the virtual reality, the scene of the accident to get attention control points around the regional coordinate data, this paper proposes an improved 2D-DLT method based on the fusion of real and virtual target. Through field test virtual calibration object to establish virtual reality the target fusion was investigated respectively for the size of the region, the virtual calibration area occupied by the same size when the calibration object size and contrast using a single calibration. With the real and imaginary target fusion effect on the survey results. Experimental results show that the virtual calibration area occupied by a larger measuring area average relative error value is small, the results of photogrammetry is better; in the virtual calibration object for the same size under the premise of regional, with the increases in the calibration area. To measure the regional average relative error value changes little, photogrammetric results basically unchanged; the real and imaginary target fusion than using single real calibration of photogrammetry results better. In order to verify the superiority of the improved 2D-DLT method based on the fusion of real and imaginary target. Finally, a case analysis of the application results, and it is proved that the improved 2D-DLT method the feasibility of the accident scene investigation.SUAV.SUAV photography photography can be realized conveniently from low altitude overlooking the ground is photographed image, and image resolution has meet the requirements of the relevant standards. The SUAV photography In the application of traffic accident scene survey, prohibiting the use of ordinary camera using photogrammetric 2D-DLT method to survey the traffic accident scene, shooting angle on the adverse effects of survey accuracy. For solving ill posed equations of 2D-DLT method to reduce the calculation, puts forward the world coordinate first image point measurement coordinates and pixel normalized. Then the singular value decomposition method to calculate the parameters of DLT SUAV photography 2D-DLT measurement method, and the effectiveness of the method is verified by simulation experiments. In order to provide guidance to the implementation of SUAV photography 2D-DLT measurement method, study the estimation method of the height of the photography. Finally, the real and imaginary target fusion for SUAV 2D-DLT photography measurement method, construction a complete method to survey the scene of the accident SUAV photography, design methods and processes, hardware equipment and test procedures. Through field test to verify the SUAV photography survey at the scene of accident The validity, the average error of measurement data is less than 1%, than the average error using a common digital camera using photogrammetry to reduce.SUAV photograph the accident scene investigation, can effectively solve the captured image caused by small angle correction precision is low, the coverage of the scene and small information available recovery less defects, improve the traffic accident scene survey precision. To enhance the usefulness of photogrammetry 2D-DLT method.

【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41;U491.31

【相似文献】