Computer vision developments to support circular economic activity in the built environment

建筑施工和拆除占我们废物流的40%，建筑物占我们隐含能源和资源使用量的类似数量。因此，在建筑环境中走向循环经济（封闭的材料循环，减少浪费和增加再利用）对于我们的可持续福祉是必要的。然而，要实现这一目标，必须克服重大的数字障碍，例如自动化数字孪生开发和资产表征。实现这些数字化进步需要在理解如何有效地在以下两者之间进行调解方面取得突破：（1）3D建筑模型（是抽象的），以及（2）真实的建筑及其元素在其建造和拆除周期阶段的3D扫描和感知信息。这形成了拟议计划的长期目标，其中数字世界中的规划，分析和虚拟工作将显着降低真实的建筑和解构世界中的风险（和浪费）。 要在短期内实现这一突破，需要解决与一类计算机视觉算法相关的三组问题，这些算法专注于从大型3D点云、相关传感器数据和上下文中获取有用信息。这些问题共享许多底层的数学运算和数据结构，用于拟合、特征匹配、变换、特征提取、主成分分析和有序查询。因此，在未来五年内，拟议的计划将在算法开发方面采取统一的方法：（1）用于语义丰富的多维3D扫描到3D模型转换的数据融合（适应性再利用项目规划、资产管理和选择性拆卸编程和优化所需），（2）实时构建对象检测和3D点云中的定位（对于跟踪部件、尺寸制造控制和选择性模块化组装规划是必需的），以及（3）对于3D扫描的不规则3D物体的装配和包装优化（对于降低建筑组装运输的碳足迹、利用重复使用的建筑材料进行设计和大规模打印量优化是必需的）。 最终，该计划将在计算机视觉，混合现实，严肃游戏，生成设计，人类认知，深度学习和机器人技术的相互依赖性和循环经济应用中建立新的知识。因此，不同的学员将获得知识和技能，以适应计算机视觉和建模原理和方法，为我们面临的建筑环境挑战开发新的算法课程，填补加拿大建筑，资产管理和建筑工程行业的关键技能和知识空白。实现这些目标将有助于克服建筑环境中循环经济的数字障碍，从而为所有加拿大人的财富和福祉做出贡献。

Building construction and demolition comprise 40% of our waste stream, and buildings account for similar amounts of our embodied energy and resource usage. Thus, moving toward a circular economy (closed materials loops, reduced waste and increased reuse) in the built environment is necessary for our sustained well being. However, significant digital barriers must be overcome to meet this goal, such as automated digital twin development and asset characterization. Enabling these digital advances requires breakthroughs in understanding how to effectively mediate between: (1) 3D building models (which are abstractions), and (2) the 3D scanned and sensed information of real buildings and their elements at stages in their construction and deconstruction cycles. This forms the long-term goal of the proposed program, in which planning, analysis and virtual work in the digital world will significantly reduce risk (and waste) in the real world of construction and deconstruction. Facilitating this breakthrough in the near term requires solving three sets of problems related to a class of computer vision algorithms that focus on deriving useful information from large 3D point clouds, related sensor data, and context. These problems share many underlying mathematical operations and data structures for fitting, feature matching, transformations, feature extraction, principle component analysis and ordered queries. Thus, over the next five years, the proposed program will pursue a unified approach in algorithm development for: (1) data fusion for semantically enriched, multidimensional 3Dscan-to-3Dmodel transformations (required for adaptive reuse project planning, asset management, and selective disassembly programing and optimization), (2) real-time built-object detection and locating in 3D point clouds (required for tracking components, dimensional fabrication control, and selective modular assembly planning), and (3) fitting and packing optimization for 3D scanned, irregular 3D objects (required for lowering carbon footprint of construction assembly shipments, designing with reused construction materials, and large scale printing volume optimization). Ultimately, the program will build new knowledge in the interdependencies among and circular economic applications of computer vision, mixed reality, serious gaming, generative design, human cognition, deep learning and robotics. Diverse trainees will thus acquire the knowledge and skills to adapt computer vision and modeling principles and methods to develop new algorithm classes for the built environment challenges ahead of us, filling critical skills and knowledge gaps in the Canadian construction, asset management and architectural engineering industries. Achieving these objectives will contribute to overcoming the digital barriers to a circular economy within the built environment, thus contributing to the wealth and well being of all Canadians.