Salman Ali Presentation

Registration of Airborne LiDAR and Aerial Images for Object Extraction

Salman Ali, Kevin Mooney

Department of Spatial Information Sciences, The Dublin Institute of Technology; engineer790@yahoo.com

LiDAR and aerial images alone or together have been used extensively in recent years for object extraction for 3D city modelling, noise modelling, land cover classification and other applications. For high accuracy and robustness in object extraction, it is essential that the two data sets have correct registration. Modern digital line scanners such as the ADS 40 from Leica Geosystems can capture multispectral images with a very high ground resolution. These line scanners rely on direct geo-referencing (GPS/INS) and, due to their design, cannot be geo-referenced accurately using ground control points alone. On the other hand LiDAR systems also use GPS/INS data for the registration of the generated point cloud. Aerial images have a higher planimetric accuracy than LiDAR and can be used as a reference for the registration of both data sets (co-registration). Co-registration, as implemented by the authors, is achieved by generating digital surface models (DSM) of the same area from each sensor and using an image matching approach to identify common points of detail, thereby allowing the lack of correspondence of registration to be computed using a 2D linear conformal transformation.

This presentation will evaluate the use of cross-correlation and adaptive Least Squares matching techniques to identify corresponding image pixels in each DSM. High correlation exists over buildings and open areas but not in areas of vegetation. The quality of the DSM generated from the aerial images is crucial and can have a very significant influence on the ability to successfully match the two DSMs.

The above methods have been successfully applied to data sets obtained from Ordnance Survey Ireland as part of a current DIT PhD study and EuroSDR data acquired specifically to assess methods and their quality currently available for the registration of both data sources.

Access presentation here (2 meg)

Return to Presentation Programme Page