Accuracy Evaluation of Airborne and Terrestrial Point Cloud Registration

Terrestrial laser scanners (TLS) and drones have revolutionized mapping in recent years. However, neither can fully cover large sites since scanners miss objects’ topmost parts and drones fail to capture objects’ vertical elements. Therefore, there is a need to integrate their data to provide complete coverage of sites. This study aims to inspect the accuracy of combining such terrestrial and airborne point clouds. To this end, data of three sites were collected using a TLS and a drone and processed separately. Airborne data were georeferenced using ground control points, but terrestrial data were not georeferenced intentionally so that it could be scaled, shifted, and rotated freely in 3D space to fit the airborne data more closely. Then, nonlinear least squares adjustment was applied for this coordinate transformation, and the results were assessed both visually and quantitatively. No major gap or misalignment between two data types could be visually identified after integration. For quantitative analysis, the root mean square error (RMSE), the change in flatness and orientation of some areas after data integration were computed. The results indicate the maximum RMSE, flatness, and orientation change were 0.037 m, 0.01 m, and 37 arcminutes, respectively, suggesting that integration error was within the allowable tolerance. Statistical significance testing was also applied to the obtained results indicating, with a 99% significance level, that the average flatness did not change after data integration, and the average orientation change was less than 15 arcminutes for all test areas.