The present study proposes a workflow to extract vegetation height for urban areas from Pléiades stereo and tri-stereo satellite imagery. The workflow was applied on a stereo image pair for Szeged, Hungary and on tri-stereo imagery for Salzburg, Austria. Digital surface models (DSMs) of the study areas were computed using the semi-global matching algorithm. Normalised digital surface models (nDSMs) were then generated. Objects of vegetation and non-vegetation were delineated based on the spectral information of the multispectral images by applying multi-resolution segmentation and support vector machine classifier. Mean object height values were then computed from the overlaid pixels of the nDSMs and assigned to the objects. Finally, the delineated vegetation was classified into six vegetation height classes based on their assigned height values by using hierarchical classification. The vegetation discrimination resulted in very high accuracy, while the vegetation height extraction was moderately accurate. The results of the vegetation height extraction provided a vertical stratification of the vegetation in the two study areas which is readily applicable for decision support purposes. The elaborated workflow will contribute to a green monitoring and valuation strategy and provide input data for an urban green accessibility study.
Keywords: 3D extraction of urban vegetation; Pléiades (tri-)stereo imagery; digital surface model; semi-global matching; support vector machine.