Towards Extracting 3-D Structural Representations of AGR Core Fuel Channels from 2-D In-Core Inspection Videos

Remote Visual Inspection (RVI) of Advanced Gas-cooled Reactor (AGR) nuclear power stations allows engineers to gain an understanding of the AGR graphite core health by investigating the incorporated fuel channels. During planned, periodic outages, video footage of the pre-selected fuel channels is acquired using specialist inspection tools and is subsequently taken offline for further analysis using visualization techniques. Current methods of visualization however provide limited structural information due to the loss of depth information as a direct result of the image acquisition process. This paper introduces a new bespoke 3-D reconstruction framework to recover lost depth information to produce 3-D point cloud reconstructions of fuel channels from inspection videos. We also present here a new, lab based, experimental rig setup with which we effectively captured data under lab controlled conditions to verify our 3-D reconstruction algorithms. Our proposed method is tested on 2-D in-core inspection videos in addition to the footage captured within laboratory conditions and outperforms state-of-the-art incremental reconstruction frameworks whilst producing a more representative 3-D point cloud for improved in-core visualization. Keyword: Advanced Gas-cooled Reactor, Visual Inspection, Structure-from-Motion