TESTFIELD TRENTO: GEOMETRIC EVALUATION OF VERY HIGH RESOLUTION SATELLITE IMAGERY G. AGUGIAROa, D. POLIb, F. REMONDINOa, 3DOM, 3D Optical Metrology Unit Bruno Kessler Foundation, Trento, Italy a b Vermessung AVT ZT-GmbH, Austria Melbourne, 28 August 2012 Motivation VHR optical satellite images are used increasingly in the scientific and civil community (e.g. for settlements analysis, disaster risk assessment, postdisaster needs assessment, etc.) DSMs are a crucial for the extraction of 3D information (e.g. building and city models, terrain geomorphologic parameters, canopy height, geologic mass monitoring, etc.) 3DOM is building a testfield over the city of Trento (Northern Italy), which comprises also 2 stereo-pairs (one GeoEye-1 and one WorldView-2) This talk presents the testfield and focuses on 3D data extraction from satellite imagery. 1
Area extents: ca. 15x15 km Min height: 150 m a.s.l. Max height: 2181 m a.s.l. Trento Population: 115000 inh. Height: ca. 200 m a.s.l. Spaceborne datasets: Sensor Acquisition date Bands Quickbird 30 Oct 2005 MS4 Quickbird 17 Jul 2006 SPOT5-VEG SPOT5-HRG 31 Aug 2006 31 Aug 2006 WV2 22 Aug 2010 WV2 22 Aug 2010 GE1 28 Sep 2011 GE1 28 Sep 2011 MS4 MS4 MS8 MS8 MS4 MS4 Avg. GSD [m] 0.6 2.4 0.6 2.4 10.0 2.5 0.51 2.0 0.51 2.0 0.50 2.0 0.50 2.0 Size x [km] Size y [km] Licensee 14,75 7,51 Univ. Trento 15,08 9,20 Univ. Trento 36,28 36,22 36,13 36,09 Univ. Trento Univ. Trento 17,64 17,64 FBK 17,64 17,64 FBK 10,00 10,00 FBK 10,00 10,00 FBK 2
Other datasets: A raster-based DSM/DTM, derived from a flight in 2006/7 at ca. 1.3 points/m2. Cell size 1 m (2 m on mountainous areas). σz,dsm=15 cm, σ z,dtm=30 cm. Reference dataset for the geometric analyses. 230 digital orthophotos (2006 and 2009), GSD 50 cm. 10 nadir aerial images (2009), acquired by RMKTOP15 analogic camera over the city of Trento (overlap: 60% along track, 30% across track). GSD 12 cm. 180 UAV images (2012), acquired by a Microdrones MD4-200 mounting a Pentax Optio A40 digital camera (overlap: 80% along track, 40% across track). GSD 3 cm 70 ground points (2011), acq. by GNSS, sub-decimetre accuracy 12 ground points (2012), acq. by total station, cm accuracy Space- and airborne imagery overview WorldView-2 GeoEye-1 Quickbird, Quickbird Aerial images (white) UAV images SPOT (not depicted, larger area) 3
GCPs WorldView-2 (WV2) stereo-pair: Acquisition on: 22 Aug 2010, 10:40 am (GMT) + MS8 (R, G, B, NIR, coastal, yellow, red edge, and NIR-2), avg GSD 0.5 m (), 2 m (MS) Extents: 17.64x17.64 km, overlap 100% Processing level: Stereo 1B Provided with RPCs FS, in-track view. angle 15.9, RS, -14.0 Cloud cover <10% GeoEye-1 (GE1) stereo-pair: Acquisition on: 8 Sep 2011, 10:20 am (GMT) + MS4 avg GSD 0.5 m (), 2 m (MS) Extents: 10x10 km, overlap 100% Processing level: GeoStereo Provided with RPCs RS, in-track view. angle 15, RS, -20 Cloud cover <4% 4
Example of artefacts found in WV-2 images Examples of artefacts found in GE-1 images Methodology General workflow: Stereo image acquisition Image orientation Image matching DSM generation Photogrammetry Orthorectification DTM extraction Building heights Building extraction Visualisation 5
Methodology Image orientation and DSM extraction Carried out within Sat-PP (4Dixplorer) Image orientation by RPCs + GCPs (sub-pixel accuracy) DSM generated a 1 m (GSD x2) Few tie points manually measured at height discontinuities No further editing/filtering In general: DSM succesfully generated from both stereo pairs In mountain areas: valleys, mountain sides and ridges well modelled In urban areas: building agglomerations, blocks with different heights, road network, some infrastructures (i.e. bridges), and rivers are well outlined. In rural areas: adjacent fields, vegetation, and buildings are recognisable on flat or hilly terrain Wrong DSM extraction due to clouds in the lower bottom part Results DSM from GeoEye-1 6
Results DSM quality assessment (preliminary results) Both DSMs compared with the DSM 3 areas (Area1, Area2, Area3) Compute height differences and statistics within areas (min, max, avg, stddev, RMSE) Area1 Historical city centre. Completely built up area, irregular building shapes, narrow streets, nearly no vegetation Area2 Railway station area. Large regular buildings, larger streets with some trees Area3 Residential area. Regular geometries, courtyards and streets with trees. Results: Area1 GE1 (all values in metres) min: -44.4, max: 26.5 sigma: 6.6 avg: 2.3 RMSE: 6.9 Error distribution: GE1, WV2 WV2 min: -53.9, max: 26.7 avg: 1.9 sigma: 7.1 RMSE: 7.4 7
Results: Area1 Profiles: Lidar, GE1, WV2 Comments: Roof height estimated quite well Errors between buildings: narrow streets not visible in the stereo-pairs due to shadows or occlusions Two churches not reconstructed: homogeneous roof material? GE1 DSM slightly more accurate than the WV2 one Results: Area2 GE1 Error distribution: GE1, WV2 WV2 (all values in metres) min: -28.1, max: 53.5 sigma: 6.9 avg: 1.7 RMSE: 7.1 min: -39.0, max: 58.4 avg: 0.7 sigma: 7.1 RMSE: 7.4 8
Results: Area2 Profiles: Lidar, GE1, WV2 Comments: General agreement between and GE1/WV2 Some differences due to vegetation One building demolished GE1 DSM slightly more accurate than the WV2 one Results: Area3 GE1 Error distribution: GE1, WV2 WV2 (all values in metres) min: -46.3, max: 28.8 sigma: 7.4 avg: 0.5 RMSE: 7.9 min: -43.7, max: 27.7 avg: 0.3 sigma: 8.5 RMSE: 8.5 9
Results: Area3 Profiles: Lidar, GE1, WV2 Comments: General agreement between and GE1/WV2 Significant height differences between tall buildings GE1 DSM slightly more accurate than the WV2 one Results: Area3 Example of occlusion between tall buildings in GE1 stereo-pair: forward view [left] and backward view [right]. 10
Conclusions and outlook Conclusions: : data acquisition completed Wrt. VHR spaceborne optical imagery, preliminary results show that: GE1/WV2 DSMs model flat, hilly and mountainous areas well (cliffs, ridges and roads are well defined) In urban areas, building blocks are recognised in the historic city centre, individual buildings are visible in residential and industrial areas Problematic areas: Narrow streets, tall nearby buildings Homogeneous/poor texture (shadows, roof covers) Presence of vegetation On the three test areas: the performance of GE1 and WV2 sensors is similar Outlook: Wrt. GE1/WV2 imagery, more extensive tests are planned Radiometric investigations DSM generation with other packages More accurate DSM analyses (e.g. masking out temporal changes) Thank you for your attention Contact: Giorgio Agugiaro Bruno Kessler Foundation 3DOM Via Sommarive, 18-38123 Trento - Italy E-mail: agugiaro@fbk.eu - Tel. +39 0461 314913 Credits: Daniela Poli, Vermessung AVT ZT-GmbH, Austria Fabio Remondino, FBK-3DOM, Italy Provincia Autonoma di Trento, Italy 3M project (co-founded Marie-Curie Actions FP7 PCOFOUND GA-2008-226070, acronym Trentino Project ) 11