【计算机科学】【2013.09】激光雷达点云与立体图像点云的融合

本文为美国海军研究生院（作者：Paul L. Basgall）的硕士论文，共80页。

激光雷达（LiDAR）点云采集技术的出现大大提高了精确、精细、三维细节模拟世界的能力。当与同样精确的摄影测量图像和摄影测量导出的点云数据相结合时，可以构建融合的数据集，从而提高精确建模和解释的潜力。

本研究的目的是演示精确的、基础的方法融合激光雷达数据和摄影测量图像及其变化检测的潜力。该项目的范围是研究光学图像与激光雷达的配准方法，重点是针对几种不同的图像类型和不同的激光雷达点云密度。建立了一种新颖的光学图像到激光雷达数据的配准方法。对于一种图像类型，使用全景数学模型的有理多项式系数（RPC）表示证明了这种方法，将精度从1.9m提高到0.5m均方根（RMS）误差。使用90%置信区间将立体图像点云数据与激光雷达点云进行比较，突出显示了包括小尺度（<50cm）、传感器相关变化和大尺度、新住宅建设在内的变化。本研究还提出了一种融合激光雷达和立体图像基层作为激光雷达/图像进一步融合的基础。

The advent of Light Detection and Ranging(LiDAR) point cloud collection has significantly improved the ability to modelthe world in precise, fine, three dimensional details. When coupled withequally accurate photogrammetric imagery and photogrammetrically derived pointcloud data, fused data sets can be constructed that improve potential forprecise modeling and interpretation. The objective of this research was todemonstrate accurate, foundation methods for fusing LiDAR data andphotogrammetric imagery and their potential for change detection. The scope ofthe project was to investigate optical image to LiDAR registration methods,focusing on several dissimilar image types and varying LiDAR point densities.An innovative optical image to LiDAR data registration process was established.This approach was demonstrated for one image type using the rational polynomialcoefficients (RPC) representation of the panoramic math model improvingaccuracy from 1.9 m to 0.5 m root mean square (RMS) error. Comparison of stereoimagery point cloud data to the LiDAR point cloud using a 90% confidenceinterval highlighted changes that included small scale (< 50cm), sensordependent change and large scale, new home construction change. This researchalso proposed a fused LiDAR and stereo image base layer as the foundation forfurther LiDAR/image fusion.

1. 引言
2. 项目背景
3. 研究目标
4. 数据处理方法、结果与分析
5. 总结
6. 结论

更多精彩文章请关注公众号：