To discover the possible solutions of error factors, this work categorizes error fac-tors into three classes – ’method limitation’, ‘data limitation’ and ‘data quality issue’.
If error factors can be solved by modifying or extending the proposed method, they belong to ‘method limitation’. Error factors influenced by data characteristics belong to ‘data limitation’. Errors that can be reduced by improving data quality are catego-rized as a ‘data quality issue’. Table 9 shows the classification of error factors. Table 10 depicts the number of each error factor cases.
In total, 38%, 32% and 30% of errors fall into the ‘method limitation’, ‘data lim-itation’ and ‘data quality issue’ categories, respectively. The ‘method limlim-itation’ and
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
Table7.Errorfactorsinbuildingregiongeneration:omissionerrors. Number TypeofcasesDescription 8.Vegetation- occluded elements
30Sincepointsinvegetationareasareremoved,buildingsoccludedbyvegetation cannotbefound.Figure15(a)depictsanexample 9.Buildingwithfew LIDARpoints14Smallelements,elementspartlyoccludedbyvegetationorlowLIDARpoint densitiesresultininsufficientinformationforgeneratingbuildingregions. Figure15(b)depictsanexample 10.Grounddetection errors6TheLIDARpointscanpassthroughsomeroofmaterialsandrecordtheheightsof innerstructures,whichcancauseerrorsduringgrounddetection.Figure15(c) depictsanexample 11.LackofLIDAR points6Theocclusioneffectorabsorptionofenergybyroofmaterialscanresultinthelossof LIDARpoints.Inthiscase,newbuildingregionscannotbegenerated. Figure15(d)depictsanexample 12.Vegetation detectionerrorin occludedareas
3Sincenospectralinformationcanbeutilizedforvegetationdetectioninoccluded areasofaerialimages,thisworkusestexturesofnDSMinclassificationto compensatefortheseareas.However,theclassificationresultisnotasgoodasthat usingNDVI.Therefore,someincorrectlydetectedvegetationareasresultinthe removalofadditionalLIDARpoints Note:LIDAR,lightdetectionandranging;NDVI,normalizeddifferencevegetationindex;nDSM,normalizeddigitalsurfacemodel.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
(a) (b)
(c) (d)
Figure 15. Omission error examples in building region generation. (a) Vegetation occluded.
(b) Few light detection and ranging (LIDAR) points. (c) Ground detection error. (d) Lack of LIDAR points.
‘data quality issue’ errors are much easier to resolve than ‘data limitation’ errors. If both the method and data quality are improved, the number of errors can be reduced by at most 68%.
6. Conclusions
For building model revision and land survey, this work proposed a new scheme that detects changes in old building models with new LIDAR point clouds and aerial imagery. With the proposed method, multiple change types can be gener-ated to reduce human determination work while surveying land. Additionally, the new epoch building regions generated provide good initial values for building model revisions.
The overall accuracy of change type determination was as high as 85%. The influ-ence of ground and vegetation during change detection can be decreased by integrating spatial and spectral information. To generate building regions, the TIN method per-forms best out of the four methods with this test data, with an overall accuracy as high as 96% in pixel-based validation. However, vegetation-occluded and non-building objects are primary sources of errors during region generation.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
Number
Type of cases Description
13. Non-building objects
27 Some man-made non-building structures are misrecognized as buildings, such as containers, trucks, high-voltage towers and carports
14. Ground detection errors
9 This error typically occurs in areas with parking ramps or revetments
15. Close buildings 3 If buildings are closely packed, the generated region may combine buildings
16. Non-building points around buildings
2 If non-building points near a building are missed during the removal process, the generated building region may be enlarged slightly
17. Occlusion detection errors
2 Since this work uses DSM in image true orthorectification and DSM cannot delineate building boundaries well, the true orthorectification result may have some errors around building boundaries. Normally, when this error occurs at a ground area, the LIDAR points can still be removed using the ground index map. However, when this error occurs in areas with vegetation, LIDAR points remain and are misrecognized as buildings. Figure 16 depicts an example
18. Vegetation detection errors
1 Some vegetation cannot be detected by NDVI and cause vegetation detection error. These vegetation areas will be misrecognized as buildings
Note: LIDAR, light detection and ranging; NDVI, normalized difference vegetation index;
DSM, digital surface model.
(a) (b)
Figure 16. An example of ‘occlusion detection errors’. (a) Aerial image. (b) Input data for region generation.
Through careful observation and classification of error factors, this work elucidates the problems that may be encountered during change detection of building models and region generation from discrete points. Besides the aforementioned solutions for method limitation, a possible improvement for proposed change detection scheme is reducing the sensitivity of rule-based threshold by integrating with probability theory
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
Table 9. Classification of error factors.
Error
factor∗ Class Reasons
1. ML The high wall points can be deleted during the wall point removal process
2. DQI These errors can be resolved when building models can delineate real roofs in an acceptable accuracy. Conversely, the proposed scheme may be used for assessing building model completeness
3. DQI This error is due to the LIDAR point density problem. With increased point density, this error factor can be resolved 4. DL If a building is occluded by another object, no data can be used for
detecting changes. This occlusion typically occurs in remote sensing and photogrammetry fields
5. DL This error occurs for occluded areas in a LIDAR point cloud or un-reflecting infrared roofs
6. ML The proposed scheme does not consider building elements
representing courtyards. Although this error is not easily resolved, courtyards are rare
7. ML This error occurs because the proposed scheme uses height difference between two epochs as one of the principal information for change detection. This error may be resolved by applying edge detection on images to match with building model edge. In addition, if the original building model contains texture information, image correlation between two epochs can be applied
8. DL The reason for this error is the same as that for error type 4—partly vegetation-occluded elements error
9. DQI The reason for this error is the same as that for error type 3—small elements
10. DL The reason for this error is that the LIDAR points pass through some roof materials
11. DL The reason for this error is the same as that for error type 5—lack of LIDAR points
12. DQI Since the outcome of using nDSM textures for detecting vegetation is poor, if occluded areas in images when taking aerial photos can be reduced in size, the incidence of this error can be reduced
13. ML This error can be resolved by differentiating building and non-building objects with additional rules or information
14. ML This error is caused by ramps. Although this error is closely related to the ground definition, this error may be solved by modifying the DEM generation method
15. ML and
DQI
Although difficult, this error may be solved by choosing a relatively better grouping threshold or method before generating building regions. Furthermore, higher LIDAR point density also allow better grouping
16. ML Although this error is usually caused by vegetation detection error or occlusion detection error, it may be solved by regularizing the generated building boundary
17. DL Since DSMs cannot delineate building boundaries well, using DSMs in true orthorectification has poor accuracy
18. DL This error occurs because some vegetation cannot be found by vegetation indices
Notes: ML, method limitation; DQI, data quality issue; DL, data limitation; nDSM, normal-ized digital surface model; DEM, digital elevation model; LIDAR, light detection and ranging;
DSM, digital surface model.
∗The error factors referred to are listed in tables 6–8.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
Class Error factors Number Total Limitation
of method
1. High wall points 6. Courtyard
7. Changed element with a small height difference 13. Non-building objects
9. Building with few LIDAR points
12. Vegetation detection error in occluded areas 15. Close buildings
15121433 47
Note: LIDAR, light detection and ranging.
(e.g. fuzzy set theory). Therefore, for elements that cannot clearly fit into any change type, further determination with including more information (e.g. line features from imagery) could be helpful.
Acknowledgement
The authors thank National Science Council and Ministry of the Interior of Taiwan for their support.
References
BOUZIANI, M., GOITA, K. and HE, D.C., 2007, Change detection of buildings in urban environ-ment from high spatial resolution satellite images using existing cartographic data and prior knowledge. In International Geoscience and Remote Sensing Symposium, 23–28 July 2007, Barcelona (Piscataway, NJ: IEEE International) pp. 2581–2584.
CHAMPION, N., MATIKAINEN, L., ROTTENSTEINER, F., LIANG, X. and HYYPPA, J., 2008, A test of 2D building change detection methods: comparison, evaluation and perspectives.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 37, pp. 297–304.
COHEN, J., 1960, A coefficient of agreement for nominal scales. Educational and Psychological Measurement, 20, pp. 37–46.
CONGALTON, R.G., 1991, A review of assessing the accuracy of classifications of remotely sensed data. Remote Sensing of Environment, 37, pp. 35–46.
CURRAN, P.J., 1988, The semivariogram in remote sensing: an introduction. Remote Sensing of Environment, 24, pp. 493–507.
DANAHY, J, 1999, Visualization data needs in urban environmental planning and design. In Proceedings of Photogrammetric Week, 22–26 September 1999, Stuttgart (Heidelberg:
Wichmann Verlag), pp. 351–365.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
GIRARDEAU-MONTAUT, D., ROUX, M., MARC, R. and THIBAULT, G., 2005, Change detection on points cloud data acquired with a ground laser scanner. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 36, pp. 30–35.
HATGER, C. and BRENNER, C., 2003, Extraction of road geometry parameters from laser scanning and existing databases. In ISPRS Working Group III/3 Workshop ‘3-D Reconstruction from Airborne Laserscanner and InSAR Data’, 8–10 October 2003, Dresden, Germany (Dresden: Institute of Photogrammetry and Remote Sensing, Dresden University of Technology), pp. 225–230.
HUBER, M., SCHICKLER, W., HINZ, S. and BAUMGARTNER, A., 2003, Fusion of LIDAR data and aerial imagery for automatic reconstruction of building surfaces. In Proceedings of 2nd GRSS/ISPRS Joint Workshop on Data Fusion and Remote Sensing over Urban Areas, 22–23 May 2003, Berlin (Piscataway, NJ: IEEE press), pp. 82–86.
IM, J., JENSEN, J.R. and TULLIS, J.A., 2008, Object-based change detection using correlation image analysis and image segmentation. International Journal of Remote Sensing, 29, pp. 399–423.
JUNG, F., 2004, Detecting building changes from multitemporal aerial stereopairs. ISPRS Journal of Photogrammetry and Remote Sensing, 58, pp. 187–201.
KNUDSEN, T. and OLSEN, B.P., 2003, Automated change detection for updates of digital map databases. Photogrammetric Engineering and Remote Sensing, 69, pp. 1289–1296.
LEICA-GEOSYSTEMS, 2006, Leica Geosystems brochure of ALS50-II. Available online at: http://
gi.leica-geosystems.com/default.aspx (accessed 10 May 2010).
LI, P., XU, H. and GUO, J., 2010, Urban building damage detection from very high resolution imagery using OCSVM and spatial features. International Journal of Remote Sensing, 31, pp. 3393–3409.
MASAHARU, H. and HASEGAWA, H., 2000, Three-dimensional city modeling from laser scanner data by extracting building polygons using region segmentation method. International Archives of Photogrammetry and Remote Sensing, 33, pp. 556–562.
MATIKAINEN, L., HYYPPÄ, J. and HYYPPÄ, H., 2003, Automatic detection of buildings from laser scanner data for map updating. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 34, pp. 218–224.
MATIKAINEN, L., HYYPPÄ, J. and HYYPPÄ, H., 2004, Automatic detection of changes from laser scanner and aerial image data for updating building maps. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 35, pp. 434–439.
MURAKAMI, H., NAKAGAWA, K., HASEGAWA, H., SHIBATA, T. and IWANAMI, E., 1999, Change detection of buildings using an airborne laser scanner. ISPRS Journal of Photogrammetry and Remote Sensing, 54, pp. 148–152.
RAU, J.Y., 2002, Geometrical building modeling and its application to the ortho-rectification for aerial images. PhD thesis, National Central University, Taiwan.
ROTTENSTEINER, F. and BRIESE, C., 2002, A new method for building extraction in urban areas from high-resolution LIDAR data. International Archives of Photogrammetry and Remote Sensing, 34, pp. 295–301.
ROTTENSTEINER, F., TRINDER, J., CLODE, S. and KUBIK, K., 2007, Building detection by fusion of airborne laser scanner data and multi-spectral images: performance evalua-tion and sensitivity analysis. ISPRS Journal of Photogrammetry and Remote Sensing, 62, pp. 135–149.
SAMPATH, A. and SHAN, J., 2007, Building boundary tracing and regularization from air-borne LIDAR point clouds. Photogrammetric Engineering and Remote Sensing, 73, pp. 805–812.
SIEBE, E. and BUNING, U., 1997, Application of digital photogrammetric products for cellular radio network planning. In Proceedings of Photogrammetric Week, 22–26 September 1997, Karlsruhe, Germany (Heidelberg: Wichmann Verlag), pp. 159–164.
SUVEG, I. and VOSSELMAN, G., 2004, Reconstruction of 3D building models from aerial images and maps. ISPRS Journal of Photogrammetry and Remote Sensing, 58, pp. 202–224.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014
TEO, T.A., CHEN, L.C., RAU, J.Y. and CHEN, S.J., 2007, Building reconstruction using a split-shape-merge method. Asian Journal of Geoinformatics, 7, pp. 31–34.
TERRASOLID, 2010, TerraScan user’s guide. Available online at: http://www.terrasolid.fi/system/
files/tscan_1.pdf (accessed 25 October 2010).
VÖGTLE, T. and STEINLE, E., 2004, Detection and recognition of changes in building geometry derived from multitemporal laser scanning data. International Archives of Photogrammetry and Remote Sensing, 35, pp. 428–433.
VOLZ, S. and KLINEC, D., 1999, Nexus: the development of a platform for location aware appli-cation. In Proceedings of 3rd Turkish-German Joint Geodetic Days, Vol. 2, 1–4 June 1999, Istanbul (Istanbul: Istanbul Technical University), pp. 599–608.
VU, T.T., MATSUOKA, M. and YAMAZAKI, F., 2004, LIDAR-based change detection of buildings in dense urban areas. IEEE International Geoscience and Remote Sensing Symposium, 5, pp. 3413–3416.
WALTER, V., 2004, Object-based classification of remote sensing data for change detection.
ISPRS Journal of Photogrammetry and Remote Sensing, 58, pp. 225–238.
WANG, Z. and SCHENK, T., 2000, Building extraction and reconstruction from lidar data.
International Archives of Photogrammetry and Remote Sensing, 33, Amsterdam, The Netherlands, unpaginated CD-ROM.
XU, J.Z., WAN, Y.C. and YAO, F., 2010, A method of 3D building boundary extraction from airborne LIDAR points cloud. Photonics and Optoelectronics, pp. 1–4.
ZHANG, L. and GRUEN, A., 2006, Multi-image matching for DSM generation from IKONOS imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 60, pp. 195–211.
ZHU, L., SHIMAMURA, H., TACHIBANA, K., LIU, Z. and GONG, P., 2009, Building change detection using aerial images and existing 3D data. Urban Remote Sensing Event, pp. 1–5.
Downloaded by [National Chiao Tung University ] at 15:19 28 April 2014