The role of control network in the accuracy of underground laser scanning surveys

Abstract

Terrestrial laser scanning is used in various fields with numerous applications, one being the documentation of heritage sites. Often scans will be georeferenced with respect to a real-world coordinate system. This is done using either direct or indirect georeferencing techniques. The indirect georeferencing method was used in this research, which uses coordinated targets, referred to as Ground Control Points (GCPs), that are captured in the scan scene. Manufacturers suggest a minimum of three GCPs are used, should the z-axis not be vertical, with additional GCPs for redundancy. Ideally GCPs should be placed evenly around the extent of the scan scene. For heritage site documentation, this is not always feasible given the unique and complex nature of each site. This research investigates the quantity and spatial variability of GCPs used in a scan scene, and the subsequent effect on the point cloud accuracy. A control test network was established at the School of Surveying (SoS), where variations of GCP scenarios were investigated, which was then applied to a case study in Arras, France. The case study being a network of underground World War One tunnels that were excavated by the New Zealand Engineering Tunnelling Company (NZETC), known as the Ronville Tunnels. The results from the SoS test network show, in this particular instance, that there is little benefit in using additional GCPs in a scan scene, should the minimum (three) be placed evenly around the extent of the area being captured. Low redundancy solutions may reduce the accuracy and robustness of georeferencing solutions, as seen with the case where large errors were present where the minimum number of GCPs were used. Geometrically poor placement of GCPs shows an increased correlation between the check points root mean square errors and range from the GCP centroid. The case study results, where the scan scene extents were hundreds of metres long, showed that it is necessary to supplement the minimum number of GCPs to mitigate uncertainties in the point cloud dataset. Similar to a least squares estimation adjustment where there are less fixed stations, there is more uncertainty for an unknown position. GCPs should therefore, be placed spatially around the extents of the scan scene and where scan scene extents increase so should the number of GCPs.