The Truth about True Orthos
In this article we discuss an innovative way to produce true-ortho imagery where all distortions are corrected and the visible ground surface is maximized.
Seeing a large area with perfectly vertical view is impossible with a frame camera or even a line scanner. Only a few pixels at the center of the capture are truly vertical, the rest have some obliqueness to the sensor. This obliqueness can be exploited to produce a digital surface model. The mostly automated digital surface model is then passed through some filtering algorithms and a bit of manual editing to remove any vertical features and man-made objects. The ortho images are then produced by correcting the distortion due to terrain using the digital terrain model. The manual editing rules vary, often they prescribe removal of bridges leading these strange scenes in Google Earth where the ortho image is draped on top of the digital terrain model. The distortions due to buildings and trees remain in this version of the ortho imagery and often lead to artifacts, especially as the resolution increases and the height of the buildings forms a significant percentage of the flying height.
The volume of imagery increases and flying heights decrease due to proliferation of UAV platforms which have a limited flying height. These platforms use wide angle lenses to maximise the footprint at these low flying altitudes, a 70° field-of-view is not atypical, leading to even more distortions due to vertical objects. These errors can typically be fixed by editing the normal orthos in image editing software, but production of true-orthos requires detail digitisation in stereo view of every vertical object – building, trees and poles.
(click picture to enlarge and use left-right arrows to move to the next picture)
An alternative is emerging to the labour intensive true-Ortho creation through the automated 3D modelling from oblique imagery combined with traditional nadir photography. Such as those collected by Microsoft Osprey, the custom Google Oblique camera or the Leica RCD30 Oblique. The automated 3D modelling with low resolution imagery and automatic radial distortion correction using in scene features can lead to soft models as seen by Google’s 3D modelling attempts. Production of true-Orthos at building scale using this technique will require higher resolution and better preservation of hard edges in man-made features such as aero3dpro models.
When producing true-orthos the top-down view may not always be appropriate in case the object of interest is roofs or the road surface. I this case a true-ortho with ray-casting from the bottom-up might be more appropriate, exposing the surfaces hidden by trees and other overhangs. While not a traditional photogrammetric product, this type of ortho can become very valuable for special interest projects. Meanwhile the asymptotic approach to the truth in true-orthos continues.