Effects of camera calibration on the accuracy of Unmanned Aerial Vehicle sensor products

Abstract

Utilization of Unmanned Aerial Vehicles (UAV) mounted with non-metric consumer grade digital cameras is on the rise globally due to their affordability and ease of operation. For high-accuracy UAV products, there is a need for accurate camera parameters determined through camera calibration. Camera calibration can be performed before (pre-calibration) or during the bundle block adjustment (self-calibration). This study aims to analyze the effect of camera calibration parameters on the accuracy of UAV products namely Digital Elevation Model (DEM) and orthoimage. Camera calibration parameters are estimated using two approaches namely self-calibration which deploys 3D image information of the scene in a bundle adjustment and a 2D reference object-based approach known as Zhang’s technique which requires image information of a planar pattern. A DJI FC220 camera mounted on a DJI Mavic Pro UAV was used. Self-calibration was deployed in Agisoft Metashape software based on Brown’s method and Zhang’s technique was deployed in MATLAB and OpenCV. Based on internal measures of accuracy, OpenCV yields the least reprojection error of 0.14 followed by MATLAB (0.79) and self-calibration (1.21). Processing without calibration yields the highest reprojection error of 2.18. Based on external measures of accuracy, that is the geometric accuracy of UAV products, self-calibration yields the least RMSE of 8.2 and 1.4 cm, for the horizontal and vertical, respectively, followed by Zhang’s technique with 9.6 and 2.3 cm in MATLAB and 13.5 and 4.3 cm in OpenCV. Processing without calibration yields the highest vertical RMSE of 20.0 and 22.9 cm for the horizontal and vertical, respectively. Comparison of accuracy of UAV mapping products computed with and without calibration emphasizes the need for camera calibration to optimize accuracy of UAV products. This study recommends assessing other photogrammetric mapping software and camera calibration approaches and the effect of flying heights on calibration parameters and mapping accuracy.