Estimating relative lidar accuracy information from overlapping flight lines

A pure statistical method for estimating relative lidar accuracy by comparing overlapping lidar datasets is described. It has been designed for commercial, fast turnaround, high-volume data production environment and has the following features. It operates on raw data and no building of grid or TIN is required. The method performs accuracy computations on statistical samples that are orders of magnitude larger than ones used in other methods reported in the literature, thus allowing detailed error analysis. Finally, the method provides a lot of relative accuracy information even when no ground control points are available. Software implementation of the method has been written and extensively tested on almost 100 Gb worth of data. Robust and fully automated, it has become an important quality control tool for data processing at TerraPoint. The method begins with the definition of a relative accuracy estimate, which is generalized from being a measure of point-to-point closeness to a measure of surface-to-surface closeness. Thus, a problem of matching the individual points in the overlapping datasets is eliminated. The generalized accuracy estimate is a function of surface size and flatness. Sampling overlap area by surfaces of different size allows one to discriminate between the random, systematic and locally systematic errors. The relative accuracy in the traditional sense of using only some tie points is a limiting case of the generalized one and can also be computed. The concepts and ideas of the method are illustrated by examples from real-life lidar projects. (C) 2002 Elsevier Science B.V All rights reserved.