University of Twente Student Theses


Manual and automated tree extraction from oblique airborne images

Rajbhandari, Ragindra Man (2012) Manual and automated tree extraction from oblique airborne images.

[img] PDF
Abstract:Trees are the essential components of urban greenness. Tree mapping has been a major concern of municipalities for estimation of urban greenness. Different remote sensing data are being used for mapping trees. Nadir view images and Airborne laser data (ALS) lack detailed information about tree profile whereas terrestrial/mobile laser scanning data, though provides sufficient information on profile of features, has limited coverage area to be used for mapping tress in urban context. Oblique aerial images provide much detailed information about profile of trees due to its oblique viewing angle and have larger coverage area compared to terrestrial equipment. The profile information of unfoliaged deciduous trees from oblique images is exploited in this research to estimate different tree parameters. Features such as geometric location, its Diameter at breast height (DBH), height are extracted manually from images with mono-plotting approach. Then an ellipse is fit to a tree crown based on its widths at different height levels. Quality of measurements in oblique imagery with this method is also assessed by GCP points and LiDAR data. These parameters are analysed to approximate threshold value to be used for automatic detection. For automatic detection of trees in oblique images, vertical lines are extracted from tree stems from multiple images of the same or different viewing directions. Base coordinates of these lines are computed by mono-plotting in object space. These coordinates are then projected into each image to identify reliable matching lines. Forward plane intersection is conducted on matched lines to generate 3D vertical lines in object space. The foot position of each 3D line is compared with mean tree coordinate measured from images. The quality of tree detection of tree is assessed manually for test site. This method is able to detect 28 out of 42 trees in the test site. The detection algorithm developed in this research shows its ability to detect most of the free standing trees, but it was unable to detect trees grouped in cluster. This is mainly due to poor extraction of lines from these trees in multiple images. Tree heights from proposed method vary much from manually extracted heights. This may due to the low contrast between a tree top and its surroundings. The result of automatic detection algorithm shows satisfactory detection of vertical lines. But the separation of vertical line from trees has not been conducted yet. Further study in this direction is expected in future. Keywords Oblique image, Mono-plotting, Multi-view, plane intersection, line extraction
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page