Rapid Three-Dimensional Shaft Scanning From Spherical Video with Photogrammetry

"This paper presents a series of case studies where spherical camera photographs were used to generate 3D shaft models using photogrammetric methods. The shaft models were, checked for accuracy by comparing them to overlapping laser scan data, provided survey data or idealized shaft models. Short lengths of model were found to more accurately reproduce the shaft surface, with a best observed accuracy of 8 mm. Longer models encountered errors with accumulated errors as exhibited by slight deviations from vertical or slightly conical model shape. At present, the models are considered acceptable for general documentation and gross change detection. INTRODUCTION Shaft inspections are required by mining regulations in many jurisdictions and are commonly completed by trained mine personnel via visual inspections. This can be time consuming and prone to errors due to lapses in attention and the rapid travel rate required to inspect an entire shaft in a reasonable amount of time. Similarly, because of the time requirements, there can be impacts on other aspects of mining as a result of occupying the shaft for inspections. Spherical cameras with supplemental light systems offer a modern addition or alternative to conventional inspections. These cameras can be lowered relatively quickly to collect high definition, fully illuminated videos of the shaft that can be reviewed in detail later while the shaft returns to service (Gillis et.al., 2019). The use of cameras offers a further opportunity for valuable survey data generation from the same original data stream. The video data can be processed into three-dimensional (3D) models using Structure-from-Motion (Westoby et al., 2012) based photogrammetric methods. Depending on the quality of the source video and availability of survey control (required for all photogrammetry models) these models can be very accurate. Shaft surveying can be conducted with traditional theodolite methods but such surveys are time consuming and limited in their data density. Similarly, scans can be completed using conventional terrestrial laser scanners, but this requires a significant number of distinct setups to provide sufficient overlap between scans. Work is being conducted both on the laser scanning (SLAM, Durrant-Whyte & Bailey, 2006) and photogrammetry fronts towards kinetic scanning (Benecke, 2017) systems. The major advancement which is permitting these kinetic systems is Inertial Measurement Units (IMUs) which can provide localization in GPS denied environments and reduce errors associated with the open traverse format associated with shaft scanning."