After
a hot shower and a hot cycle of your seagulled clothes, its time to process the
images and make a model.
Thankfully, there are some very clever people on this
planet who make normal people (like me) look very clever with minimal
understanding of the dark arts of spatial processing. The little that I do
understand is that:
- All images should be geo-referenced. This is automatic if you are using standard commercially available drones.
- There should be a good overlap (80%) of all images. If you are brave enough to send your drone into the sky on a pre-programmed flight path (easier said than done), this is simple- providing you have enough memory (you usually need many gigabytes). This does mean you get many similar photos like these filling up your hard drive:
3. You should take all the images on the same day and ideally
within a short period. This is to do with image correction and light conditions-
there is only so much that a computer can do to deal with shadows. Its quite
striking the difference in light between overcast and sunny day.
Once
you have the images, all you need is access to a programme that can stitch the
images together accounting for position, height and angle of the camera and
other visual distortions (in a process called ‘orthorectification’).
The
process involves a complex pixel-matching exercise that produces a three
dimensional cloud of points, which is called a point-cloud. The point cloud
measures the distance between each pixel so the spatial arrangement of the
pixels is known.
If
you zoom in, it looks rough and messy, but if you zoom out, it looks familiar.
The cloud of points is then processed to create a smooth digital surface model or DSM. This happens automatically once you upload your images and select the outputs. Sounds too simple? It is that simple. The trouble is that the programmes capable of undertaking these tasks requires a subscription which costs thousands of pounds. It is troublesome if you cant afford to pay for it- but it is impressive.
No comments:
Post a Comment