The advance in modern societies shape the look of megacities millions of people dwell in nowadays. To ensure the safety of human lives, urban environments, characterized by myriads of buildings and civil infrastructure, among other structures, need to be inspected and maintained regularly. However, conventional procedures in performing such tasks are labor-intensive and inefficient. With the help of advanced technology in sensing and computing, this thesis documents the design and implementation of a system that can facilitate with the tasks of building surveying that are essential in building management. The system consists of 2D and 3D vision sensors and is equipped with computing units accelerating deep learning algorithms. The ability of the system in assisting with building...[ Read more ]

The advance in modern societies shape the look of megacities millions of people dwell in nowadays. To ensure the safety of human lives, urban environments, characterized by myriads of buildings and civil infrastructure, among other structures, need to be inspected and maintained regularly. However, conventional procedures in performing such tasks are labor-intensive and inefficient. With the help of advanced technology in sensing and computing, this thesis documents the design and implementation of a system that can facilitate with the tasks of building surveying that are essential in building management. The system consists of 2D and 3D vision sensors and is equipped with computing units accelerating deep learning algorithms. The ability of the system in assisting with building management processes is demonstrated through the incorporation of a data-driven pipeline detecting building defects on concrete structures. Taking into account the more practical usage scenarios of the system, a calibration method capable of dynamically finding relative poses between a pair of sensors is documented. It separately estimating the sequential poses of different vision sensors before combining the dynamic poses with geometric markers. Through analysis, it is shown that the method can recover the dynamic relative poses between a pair of 2D and 3D sensors with accuracies. The commercial prospect of the system lies in the pipeline that it enables. In this thesis, a business development is drawn to establish market foothold by concentrating on servicing customers from specific industries before disruptively expanding to other markets.