3DReconstruction_test1

In CASA UCL we are currently looking into 3D reconstruction from drone outputs. Pix4D was originally used in the Remaplima project, to generate pointclouds of the two case studies in Lima, Peru. The innovative use of drones to capture the city on the spot and the instant generation of accurate 3D models through relevant applications, opened up new pathways in the mapping of rapidly changing urban environments.

Today there are many 3D reconstruction software out there, but it is no wonder that Pix4D now collaborates with most of the well known drone experts. It has advanced specialized features on generating 3D models from areal images with GPS coordinates. The technique is called photogrammetry and it has become extremely popular within the last year. Studies that have compared different techniques, find that photogrammetry can be as accurate as Lidar and have several competing advantages*. Moreover it is now possible to import images or videos from any camera and with very little input from the user, generate impressive 3D pointclouds, Meshes and 2D orthomosaics.

The hardware needed for a drone related project, is quite easy to get a hold of and relatively inexpensive. UAV's can now be bought on-line or can be custom-made, as with Richard's "spidey" drone. The camera can be mounted on the frame of the drone and set to take pictures periodically, as it flies over the area to be mapped. However not everywhere is allowed to fly drones. Especially in London there are very strict flight regulations. To check where you can fly drones: http://www.noflydrones.co.uk/*.

Study: To test the software before a flight, we took some photos from of the well-known physical 3D model of London currently installed at the building centre, as suggested by our professor Andy Hudson Smith. We took a total of 40 pictures of an area in Bow. The photos were uploaded into Pix4D Mapper and with draft settings, we had a complete orthomosaic of the area and a pointcloud within 15 min. Overlapping images from different angles, allowed the reconstruction of a 3D model, which originally had multiple distortions. However, it was expected considering we took standard photos using a nexus 5 with a selfie stick... The software analyses images and produces very detailed quality reports, a very useful feature considering that a high-res render might take over 10 hours. In a second test reconstruction using the same rough photos but this time using high resolution settings, the results were impressive.

Here are some tips for taking photos:

- make sure to have plenty of overlaps.
- make sure to have as less shadows as possible in objects.
- make sure the light in consistent
- make sure you don't have many moving objects in images
- make sure the height of the images is consistent.

There are several parameters which can then be adjusted to produce optimum results. Camera calibration is not one of them though. Pix4D has made sure to recognise supported cameras and use the focal length and settings of each model. Further customization from the user is possible.

What can be done with the final model? There are plenty of options in the exported outputs. I was particularly impressed with the quality of the 3D mesh and the animation feature. The software can also export into standard CAD application and all the well known 3D editing software. Looking forward for the first flight test. Watch this space.

The central London model from the Building Centre scale  1:1500 / 3D reconstruction of the central London model.

References:

Remaplima: 

http://remaplima.blogspot.co.uk/

Mapping Beyond the Palimpsest:

https://www.bartlett.ucl.ac.uk/dpu/mapping-beyond-the-palimpsest

*Baltsavias, M., 1999. A comparison between photogrammetry and laser scanning, ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 54, pp 83–94

*Beasy, C., 2007. Lidar and Photogrammetry, How Do I Choose?, Eagle Mapping, URL: http://www.eaglemapping.com/ publications/Lidar%20-%20Photo%20Article.pdf (last date accessed: 22 Dec 2015).

*Leberl, Franz, et al. "Point Clouds." Photogrammetric Engineering & Remote Sensing 76.10 (2010): 1123-1134.