A New Pipeline for Detailed Large Scale Geometry Acquisition and Analysis

人们经常将3D采集与2D采集进行比较：虽然2D采集比以往任何时候都更好、更便宜、更通用，从而产生了非常大的数据集和图像处理研究的民主化，但不幸的是，3D采集却并非如此：它仍然需要昂贵的设备、大量的后处理和操作员培训。此外，准确的大规模收购需要大量的规划、物流和自动化，目前还没有有效的解决方案。这并不是因为需求不足，因为3D采集市场预计到2023年将超过62.2亿美元，然而，尽管在几乎所有领域都有实际应用，但以高分辨率扫描大对象或区域仍然非常昂贵和耗时：AR/VR、机器人、制造、质量控制和测试、土木工程、基础设施检测、航空工业、国防、采矿、气候变化等等。在这个项目中，我们计划弥合这一差距，并为3D采集渠道提出新的方法，使难以捉摸的3D采集渠道成为主流，既包括消费者，也包括行业主流。我们从提高传感器的可靠性和准确性、将数千个几何部件缝合成一个统一的模型、使用自主代理(如无人机和/或机器人)规划和自动化3D采集等方面提出了新的技术。我们提出的方法可以对在扫描过程中发生小变形的物体进行3D采集，例如从采集过程中的人呼吸或动物移动或物体SCH，如树木被风变形。物体的延时扫描可以提供对扫描的物理结构的独特洞察，以及随着时间的推移可能对其施加的力的类型。我们提出了一种新的逆向物理框架，其中利用有限元方法，可以应用智能差分算子来检测丢失或增加的几何形状(即，物体的一部分掉落，或物体的一部分凸起)以及弹性和塑性变形方面的结构变化，并进一步预测变形轨迹。这对于基础设施维护非常有用，采购自动化和准确的分析可以帮助及早发现严重问题。除了在快速增长的市场中直接使用这些技术外，这项工作还在土木工程和采矿中有重要的人类安全应用，如民用基础设施维护、核电站内部和勘探矿山。此外，它还将产生新的研究和独特的数据集，可以进一步用于解决其他问题，因此社区和加拿大工业作为一个整体将从这项研究中受益匪浅。

3D acquisition is often compared with its 2D counterpart: while 2D acquisition is better, cheaper and more versatile than ever yielding very large datasets and a democratization of image processing research, unfortunately not the same can be said of 3D acquisition: It still requires expensive equipment, significant post-processing and operator training. Furthermore, accurate large-scale acquisition requires significant planning, logistics and automation that currently does not have efficient solutions. This is not for a lack of demand as 3D acquisition market is projected to exceed 6.22 billion USD by 2023 and yet, scanning large objects or areas with high resolution is still very expensive and time consuming despite having practical applications in nearly all fields: AR/VR, robotics, manufacturing , quality control and testing, civil engineering, infrastructure inspection, aeronautics industry, defense, mining, climate change to name a few. In this program we are planning to bridge the gap and propose new methodology for the 3D acquisition pipeline that can bring the elusive 3D acquisition pipeline into the mainstream, both consumer as well as industrial mainstream. We propose new technologies from improving sensor reliability and accuracy, for stitching together thousands of geometric pieces into one unified model, planning and automation of the 3D acquisition using autonomous agents such as drones and/or robots. We propose methods that can do 3D acquisition of objects that undergo small deformations during scanning such as from people breathing during acquisition or animals moving or objects sch as trees being deformed by wind. Time lapse scanning of objects can provide a unique insight into the physical structures scanned as well as the type of forces that might have been applied to them over time. We propose a novel reverse physics framework where using the FEM methods a smart difference operator can be applied that detects structural changes in terms of missing or added geometry (i.e. a piece of the objects fell, or a part of the object bulged) as well as in terms of elastic and plastic deformation and predict further the deformation trajectory. This can be very useful for infrastructure maintenance, where acquisition automation and accurate analysis can help detect early serious problems. In addition to the direct use of these technologies in a fast-growing market, this work has important human safety application in civil engineering and mining such as civil infrastructure maintenance, interior of nuclear power plants and exploring mines. Additionally, it will generate new research and unique datasets that can be further used to solve other problems thus the community and Canadian industry as a whole will greatly benefit from this research.