基于多模式匹配的复杂城市环境MMS精确位姿恢复方法

The construction of high precision positioning and altitude for MMS platform in complex urban environment with GNSS signal being obscured has been a hot and difficult problem in unmanned driving research. Traditional methods, such as methods using IMU and SLAM, adopt a sequential approach, which are dependent from results of previous tracks and can not solve the problem of accumulated drifting errors. The accumulated errors are extremely severe in the GNSS-denied areas, which is common in complex urban environment. Aiming at obtaining precision position and altitude information of the MMS platforms, this project integrates the widely available photorealistic meshes produced from state-of-the-art oblique photogrammetry pipeline. The involvement of the photorealistic meshes will provide additional global constraint information in the GNSS-denied areas and overcome the problem of accumulated drifting errors in traditional sequential approaches. This project consists of the following three research objectives: 1) multiple approaches for multimodal data matching, including the photorealistic meshes from oblique images, panorama and LiDAR point clouds on board MMS platform; 2) Combined bundle adjustment of multiple matching information with global constraints for the precision estimation of position and altitude of MMS platform; 3) Research and develop a prototype system and apply to a typical area, which works in conjunction with the Shenzhen Institute Center of Digital City. The output of this project will support street-level detailed reconstruction and autonomous driving in the future, which will also expand the theory and method of modern photogrammetry.

GNSS信号受遮蔽的复杂城市环境下MMS精确位姿恢复，一直是无人驾驶研究的热点与难点。传统序贯式MMS位姿估算法方法，累积误差问题突出，无法有效解决复杂城市环境GNSS信号长时间失锁条件下的位姿精度问题。本项目将针对复杂城市环境MMS精确位姿恢复瓶颈，充分利用已有的高精度倾斜摄影测量数据产品，在GNSS失锁区域，提供额外的全局约束，突破现有序贯式方法的误差累积问题，主要包含下面三个研究内容：1）实景三角网模型、车载全景影像、车载点云在内的多模态数据的多模式匹配方法；2）顾及多匹配特征全局约束的MMS高精度位姿信息恢复方法；3）设计研发原型系统，并选择深圳市典型城区进行实验分析。本项目成果将为新一代车道级道路场景重建以及无人驾驶等提供关键技术支撑，并丰富和发展摄影测量理论方法。

针对复杂城市环境MMS精确位姿恢复瓶颈，充分利用已有的高精度倾斜摄影测量数据产品，在GNSS失锁区域，较系统地突破了面向实景三角网模型、车载全景影像、车载点云在内的多模态数据的多模式匹配方法基础难题，有效提高了密集城区、西南高速公路长大隧道等场景中MMS位置信息精度。取得的主要理论方法创新成果包括：（1）基于几何图像的空地影像匹配方法；（2）复杂城市场景多模态点云配准方法；（3）长大隧道GNSS拒止环境MMS精准位姿恢复方法。共发表高水平论文20篇（其中本领域顶刊 ISPRS 会刊论文8篇）、授权发明专利1项，获得软件著作权1项；项目负责人丁雨淋获得测绘科技进步奖一等奖2项（2020，2021），詹天佑科学技术奖青年奖（2022）、全国铁路青年科技创新奖（2020）、测绘科技进步奖一等奖（2021，3/15）、西南交通大学优秀共产党员（2020）、西南交通大学科技新星（2020）等奖励与荣誉；培养研究生 12人（博士生 6 人和硕士生 6 人）。