【作者】 焦宏伟；

【导师】 秦石乔；

【作者基本信息】 国防科学技术大学 ， 光学工程， 2012， 博士

【摘要】 成像激光雷达和双CCD立体视觉是两种不同的三维测量技术，论文根据成像激光雷达距离测量精度高而空间分辨率低以及双CCD立体视觉距离测量精度受纹理、光照、远近距离等因素影响严重而空间分辨高的特点，提出将这两种技术复合的三维精细成像技术，通过对二种技术获取数据的的匹配和融合处理，得到分辨率与CCD相当而距离精度与激光雷达相当的三维精细图像。论文主要工作如下：1）对基于成像激光雷达和双目立体视觉的三维精细成像技术的相关理论和关键技术进行了研究，提出成像激光雷达与双CCD立体视觉复合的系统实现方案。研究了双CCD立体视觉模型与基本原理，对双CCD三维测量精度影响因素进行了分析，根据成像激光雷达与双CCD的视场约束、精度约束关系，优化设计了复合系统的结构参数，并建立了一套成像激光雷达与双CCD立体视觉相融合的实验研究系统，为二者复合的三维精细成像理论研究与实验验证打下了基础。2）研究了摄像机参数标定方法和双摄像机坐标系匹配标定方法。在分析现有摄像机标定算法的基础上，引入镜头切向畸变，建立更完整的畸变模型，对平面模板标定法进行了改进；进一步研究了平面模板标定法对双CCD坐标系实现匹配标定的方法。3）根据三维成像激光雷达和摄像机坐标系标定的特点，提出了一种旋转参数与平移参数分离的变换矩阵求解方法，利用这种方法，只需变换3个平面标定靶位置即可计算出二者坐标系之间的变换关系，进而设计了一种由三个平面构成的三维立体标定靶，通过采用主成分分析法对三维靶平面进行拟合，降低了平面拟合误差。该方法只需三维靶的一个位置即可完成激光雷达与摄像机坐标系之间的标定，具有精度高速度快的特点。4）研究了基于动态规划的稠密立体匹配方法和激光雷达深度信息融入图像立体匹配的方法，在此基础上提出了利用激光雷达提供的视差信息引导稠密匹配的支点法，该方法能提高立体匹配的速度和精度，对无纹理区域和遮挡区域能有效处理，解决了立体匹配中的难题。仿真结果表明融合雷达数据的动态规划方法的误匹配率要分别低于雷达数据的误匹配率和动态规划方法的误匹配率，对真实场景的双摄像机图像进行视差计算，结果表明本文方法能对图像中的遮挡区域进行有效检测，对于雷达能够提供先验视差图的区域，不论是纹理丰富区域还是无纹理区域，该方法都可以得到较好的视差结果。5）通过对多种重建方法的比较，确定了在未进行校正的双目视觉系统中三维重建使用重投影误差最小法，在经过校正后在平行双目视觉系统中直接使用三角法进行重建。通过成像激光雷达与双CCD立体视觉复合实验系统对典型场景进行实测，得到了精细的三维重建结果，验证了基于激光雷达与CCD立体视觉复合的三维精细成像技术的有效性。更多还原

【Abstract】 Imaging lidar and dual-CCD stereo vision are two different three-dimensionalmeasurement techniques, they have their owen traits, lidar measurement has the featureof high distance precision and low spatial resolution, dual-CCD stereo visionmeasurement has the feature of high spatial resolution and distance precision of whichmay be affected by texture, light, and the distance factors. According to the features ofthe two measurement techniques, the paper introduce a fine three-dimensional imagingtechnology which can fuse these two techniques, by matching and fusing of thethree-dimensional images measured by two kinds of technique, the technology can getfine three-dimensional image with spatial resolution of CCD and distance precision oflidar. The paper mainly contains the following:Firstly, the related theories and key technologies of the fine three-dimensionalimaging technology based on lidar and binocular stereo vision are researched; thecomposite system implementation of imaging lidar and dual-CCD stereo vision ispresented. The paper researchs on the model and basic principles of dual-CCD stereovision, analyses the accuracy affected factors the of dual-CCD three-dimensionalmeasurements. According to the field of view constraints and the accuracy constraintsof imaging lidar and dual-CCD, the optimal structural parameters of the compositesystem are designed and an experimental integration system of imaging lidar anddual-CCD is established, which lays the foundation for composite finethree-dimensional imaging theory and experimental verification.Secongly, the camera calibration algorithm and two camera’s coordinate systemcalibration algorithm are researched. By analysing the existing camera calibrationalgorithm and introducing the lens tangential distortion, a more completed distortionmodel is established and the planar pattern calibration method is improved; further more,a dual-CCD coordinate system calibration algorithm based on planar pattern isresearched.Thirdly, based on the features of calibration of3D imaging lidar coordinate systemand camera coordinate system, a rotation and translation parameters separated methodfor solving transformation matrix is proposed, by using this method, the transformationmatrix can be solved only through changing the planar calibration target position threetimes. And then, the paper designes a three-dimensional calibration target which ismade up of three flat planes for solving transformation matrix, by using principalcomponent analysis on the plane of the three-dimensional target, the plane fitting errorcan be reduced. This method only needs one three-dimensional target position tocomplete the calibration of the lidar and the camera coordinate system, and has thefeatures of high precision and fast speed. Fourthly, the dense stereo matching method based on dynamic programming andthe method of merging lidar depth information into the image matching are researched.Base on the research, the paper proposes a pivoting method to fusing lidar data andCCD camera data, which use lidar disparity information as pivots to guide dense stereomatching. This pivoting method can improve the speed and accuracy of stereo matchingand can effectively deal with the textureless regions and occluded regions which areproblems in stereo matching. The simulation results show that the false match rate ofthis fusion method is lower than the lidar data and dynamic programming methodrespectively. In the real scene experiments, the results show that this method can detectoccluded regions in the image effectively, and in the region of lidar priori disparity canbe provided, this method can get a better disparity result regardless of the region istextured or textureless.Finally, through the comparison of a variety of reconstruction methods, the paperdecides using the minimum re-projection error method in unrectified binocular stereovision system and using the triangulation method in rectified binocular stereo visionsystem. By measuring the typical scene using composite system of imaging lidar anddual-CCD, a fine three-dimensional reconstruction result is got, and the effectiveness offine imaging technology is verified.更多还原