Researchers build high res 3D scans from drone photos
21 March 2015
Construction of the 30-metre tall statue of Christ The Redeemer, overlooking Rio de Janeiro from a nearby mountain took nine years before it opened in 1931; it took just a few hours to build the first 3D scan of the monument from over 2,000 photos taken by a small drone circling it.
The digital replica of the monument built last month from images captured with an ordinary digital camera, is accurate to between two and five centimetres, enough to capture individual mosaic tiles.
The exercise formed part of preservation efforts for the monument and was intended to demonstrate how the use of drones could lead to a dramatic increase in high-resolution 3-D replicas of buildings, terrain, and other objects.
According to Christoph Strecha, CEO and co-founder of Pix4D, 3-D imaging of the kind could find applications in diverse fields including speeding up construction and helping Hollywood make better special effects. Strecha's company led the project in collaboration with drone manufacturer Aeryon Labs and researchers at PUC University of Rio de Janeiro.
Thanks to mapping tools built by companies including Google and Apple, making outdoor 3-D imagery has become commonplace in recent years. But it mostly makes use of 3-D data from aircraft carrying expensive equipment.
It also involves a technique called lidar, which uses lasers. The incredibly detailed imagery that gives you a feeling of actually flying over cities in Apple's ''flyover'' mode is made by processing images captured by a complex and expensive array of cameras. Highly accurate GPS technology is used in both techniques.
Drones do not have very reliable GPS fixes by comparison, and cannot carry large sensors or cameras. A fix or more accurately ''Time To First Fix'' (TTFF) is the time taken by a GPS receiver to acquire satellite signals and navigation data, and calculate a position solution (called a fix).
However, drones score when it comes to economy and Pix4D's software can build highly accurate models by comparing many different overlapping photos, says Strecha. In fact, it would not have been possible to user lidar with the Rio statue, given its size, shape, and location, he added.
Similar projects are underway elsewhere and at the Technical University of Graz, Austria, researchers led by professor Horst Bischof are developing software that extracts information from such images.
For instance, a software developed by the researchers calculates measurements necessary for producing custom-fit thermal insulation. The software is being used by a company that restores old buildings.
With the image processing more or less a solved problem, the future of drone scanning would be determined largely by how drones could be controlled or co?rdinated in challenging conditions like the winds around Rio's Christ, or to cover larger areas, according to Carl Salvaggio, a professor at Rochester Institute of Technology's Center for Imaging Science. He adds, drones are good for small-scale projects but traditional aircraft offer the time in the air to collect whole cities. He adds, perhaps when there are 'armies' of drones in the air, a different landscape may be seen to emerge.
The NEXT lab of PUC University in Rio de Janeiro had long dreamt of creating an accurate 3D model of the Christ statue, but the numerous facades and high angles for scanning the whole statue, posed challenges for conventional lidar technology. There was also enormous cost of laser scanning and lack of a highly reliable aerial method with the hardware needed to acquire the data.
Further the project called for a highly controllable and reliable UAV for the flight, one able to withstand strong wind drafts and guarantee no accidental damage to the monument, for which Pix4D and NEXT chose the small Aeryon Scout, a Vertical Take-Off and Landing (VTOL) quad rotor UAV that allowed for safe operation.
According to Strecha, creating a model for Christ in Rio was perfect because it had such detail. It was the perfect structure to show the possibilities of drone usage in image processing technology, where traditional methods like laser scanning were not practical.
The team collected 3,584 photos over six consecutive days with the Aeryon Scout, collecting what came to a total of 3,584 photos before the visitors arrived each morning. The reconstruction with PixFDmapper Pro, involved 2,090 of those photos and in addition to on-site linear measurement taken by the project team, and 82 manual tie points to merge the sub-projects.
The final 3D model emerged in complex form both as a 134.4 million point point-cloud and as a full 3D textured mesh of 2.5 million triangles.