Towards automated extraction for terrestrial laser scanning data of building components based on panorama and deep learning

Terrestrial laser scanning has been widely used in the dimensional quality assessment (DQA) on building components, but the extraction of component data is mostly done by a specialized software which is not applicable to the building components assembled in a factory. The need for manual operations makes it difficult to fulfill the notion of efficient component manufacturing. Notably, due to the successful applications in image processing, deep learning has been extended to the field of quality inspection and damage diagnosis of building components, which can quickly locate targets in images. As deep learning has excellent performance for target detection and current scanner can produce a panorama with the scan resolution, it is possible to automatically extract the targets using deep learning by correlating the panorama with the scanned data. To the best of our knowledge, little research has been on the combined use of panorama and deep learning for automated data extraction of building components. Therefore, this paper proposes an approach using the panorama and deep learning to automatically extract building component data for DQA. To determine suitable scan parameters, parametric tests are first carried out on scan distance, scan resolution, scan color and scan angle. Experimental test is then conducted on a deformed concrete-filled steel tubular column to validate the proposed approach.