$50K prize purse for solutions to advance computer vision for time-sensitive mapping
The National Geospatial-Intelligence Agency (NGA), Johns Hopkins University Applied Physics Laboratory and NASA have launched a crowdsourcing competition, the “Overhead Geopose Challenge,” which is being implemented by DrivenData and HeroX.
The challenge seeks solutions that make overhead images more useful for time-sensitive applications like security and disaster response. Participants are competing for a prize purse of $50,000. To accept the challenge, visit this website.
The challenge seeks to build computer-vision algorithms that can effectively model the height and pose of ground objects for monocular satellite images taken from oblique angles.
Overhead satellite imagery provides critical time-sensitive information for disaster response, navigation, and security. Most current methods for using aerial images assume that the images are taken from directly overhead.
However, the first images available are often oblique. These camera orientations complicate useful tasks like image alignment and change detection that are needed to ensure that maps are accurate in dynamic environments.
Solvers will transform RGB images taken from a satellite to accurately determine each object’s real-world 3D structure or “geocentric pose.” Geocentric pose is an object’s height above the ground and its orientation with respect to gravity. Calculating geocentric pose helps with detecting and classifying objects and determining accurate object boundaries.
NGA is continuously seeking novel concepts for assured positioning, navigation, and timing, said NGA Research Chief Scientist John Main. “We think the broader innovation community can help us support disaster response operations more effectively by exploring novel ways to understand where things are happening — or even better where things will be happening — during an emergency event. This is NGA Research’s goal and the reason we tap scientific and technical talent and ideas from national labs, universities, and businesses.”
“This is a hard computer vision problem that will benefit from the diverse approaches that a machine learning challenge can bring to bear,” said Greg Lipstein, co-founder and principal of DrivenData. “This is a great opportunity to bring together a fascinating dataset with a passionate community of experts to generate innovative solutions for disaster response and other time-sensitive needs”
“Our global network of problem-solvers is perfectly suited to support the advancement of state-of-the-art methods for using and understanding satellite imagery,” said Christian Cotichini, CEO, HeroX. “Having great minds come together for an initiative like this will help us more quickly and more effectively support people in times of need.”
Eligibility to Compete and Win Prizes
The challenge is open to anyone aged 18 or older not affiliated with DrivenData or the challenge sponsor and permitted to compete under the laws of the U.S. and their local jurisdictions.
Finalists will be determined by performance on a private test set of historical measurements, and bonus prizes will also be awarded for qualifying scores that reflect innovative approaches. All prize-winning approaches will be shared under an open-source license for learning and use by the community.
