Open source tool for DSMs generation from high resolution optical satellite imagery: development and testing of an OSSIM plug-in

The fully automatic generation of digital surface models (DSMs) is still an open research issue. From recent years, computer vision algorithms have been introduced in photogrammetry in order to exploit their capabilities and efficiency in three-dimensional modelling. In this article, a new tool for fully automatic DSMs generation from high resolution satellite optical imagery is presented. In particular, a new iterative approach in order to obtain the quasi-epipolar images from the original stereopairs has been defined and deployed. This approach is implemented in a new Free and Open Source Software (FOSS) named Digital Automatic Terrain Extractor (DATE) developed at the Geodesy and Geomatics Division, University of Rome ‘La Sapienza’, and conceived as an Open Source Software Image Map (OSSIM) plug-in. DATE key features include: the epipolarity achievement in the object space, thanks to the images ground projection (Ground quasi-Epipolar Imagery (GrEI)) and the coarse-to-fine pyramidal scheme adopted; the use of computer vision algorithms in order to improve the processing efficiency and make the DSMs generation process fully automatic; the free and open source aspect of the developed code. The implemented plug-in was validated through two optical datasets, GeoEye-1 and the newest Pléiades-high resolution (HR) imagery, on Trento (Northern Italy) test site. The DSMs, generated on the basis of the metadata rational polynomial coefficients only, without any ground control point, are compared to a reference lidar in areas with different land use/land cover and morphology. The results obtained thanks to the developed workflow are good in terms of statistical parameters (root mean square error around 5 m for GeoEye-1 and around 4 m for Pléiades-HR imagery) and comparable with the results obtained through different software by other authors on the same test site, whereas in terms of efficiency DATE outperforms most of the available commercial software. These first achievements indicate good potential for the developed plug-in, which in a very near future will be also upgraded for synthetic aperture radar and tri-stereo optical imagery processing.