PITTSBURGH, PENNSYLVANIA, UNITED STATES - Hanbyul Joo is working on his swing. But he’s not at a baseball park, instead he’s in a massive geodesic dome in the basement of a building at Carnegie Mellon University. As Joo swings, more than 500 cameras capture his motion on video. Combined and processed, those videos make up the elements for the most advanced 3D reconstruction ever achieved.
The two story dome is called the Panoptic Studio and its made up of 20 panels, each of which houses 24 cameras. That coupled with an additional 30 HD cameras and depth sensors add up to ensure that the system can capture more than 100,000 different points of motion at any given moment.
“One minute of videos is about 500 gigabytes which is really huge data. To handle that data we are using 120 hard drives only for the capture,” said Hanbyul Joo, a computer scientist at Carnegie Mellon University in the United States.
Thousands of cables snake around the dome feeding the video singles to a bank of computers that store the data. After the video is captured, it’s processed using algorithms which allow the scientists to turn what are essentially 500 different video feeds into a pinpoint accurate 3D reconstruction.
According to Joo, these reconstructions have a wide range of applications, the most evident being in sports where the ability to analyze and correct even the smallest motions could mean the difference between a good kick and a perfect one.
“The main purpose of this system is to analyze very interesting dynamic movements or dynamics things such as a sports players’ motion,” said Joo. “Using our system we can compare those movements so we can see how to teach the right motion to people,” he added.
But even more exciting is the technology’s potential in the fields of social and behavioral science. The studio’s ability to capture even the most subtle movements could one day allow doctors to screen for early signs of diseases ranging from dementia to autism.
The researchers are also looking to take their technology outside the lab. Joo says that, in principle, the same algorithms can be used to generate 3D reconstructions of concerts and sporting events using the media produced by thousands of people snapping selfies and making videos.