QUALITY ASSESSMENT OF IMAGE MATCHERS FOR DSM GENERATION - A COMPARATIVE STUDY BASED ON UAV IMAGES

摘要

Recently developed automatic dense image matching algorithms are now being implemented for DSM/DTM production, with their pixel-level surface generation capability offering the prospect of partially alleviating the need for manual and semi-automatic stereoscopic measurements. In this paper, five commercial/public software packages for 3D surface generation are evaluated, using 5cm GSD imagery recorded from a UAV. Generated surface models are assessed against point clouds generated from mobile LiDAR and manual stereoscopic measurements. The software packages considered are APS, MICMAC, SURE, Pix4UAV and an SGM implementation from DLR. The evaluation is conducted on a typical urban scene of 354 m × 185 m size, containing buildings, roads, variable terrain and tropical vegetation. DSMs are initially generated by the five software packages and then co-registered to the ground reference data using least-squares 3D surface matching, which minimizes the squared sum of the Euclidean differences between the matched DSM and the ground reference. The RMSEs, standard deviations and the error distributions (the analysis of blunders in particular) are used for the evaluation of the matchers on both solid (buildings, road surfaces, bare earth, etc.) and 'soft' objects (trees, bushes, etc.). The analysis covers the full dataset, the case of solid objects only, and finally a buildings-only DSM. The results of the experiments provide a useful indicator of matcher performance under the particular UAV imagery conditions considered.