Measurement accuracy on 3d point cloud generated using multispectral imagery by different calibration methods

The state-of-the-art lightweight multispectral cameras are widely used for low altitude remote sensing, also can be exploited as a tool for close-range photogrammetry application. The acquired imagery can be used for generating the 3D model using Structure-from-Motion/ Multi-view Stereo (SfM/MVS) processing software. In photogrammetry, camera calibration is an essential step for accurate measurement. The parameter of the camera system can be estimated using photogrammetric self-calibration bundle-adjustment, or by automatic and straightforward calibration procedure developed by computer vision (CV) community. When using SfM/MVS photogrammetry software, the pre-calibration value is not required, as the algorithm calculates the parameter as a part of point cloud construction process. Nevertheless, processing with the uncalibrated image is only suitable when no metric accuracy required in the modelling project. This paper aims to evaluate the measurement accuracy on generated 3D point cloud based on different estimated parameter method. The evaluation of measurement accuracy started by estimates the camera’s interior parameter using two different approaches; photogrammetric self-calibration bundle-adjustment and computer vision calibration. The estimated parameter from both methods then imported into commercial SfM/MVS software to construct the 3D point cloud. The point cloud also generated using uncalibrated images and used for measurement accuracy assessment. All parameters applied to the same datasets involved three different check-fields. Two accuracy assessments were performed by comparing the check-points and check-distance extracted with the total station measurement. As a result, the point cloud generated using photogrammetric approach provides the most accurate result on both assessments. While the automatic on-the-job self-calibration shows inconsistent results.