Accurate Uncertainty Estimation for Computer Vision Based Geo-location

ABSTRACT: As vision systems are continuously increasing their quantity and capabilities for capturing aerial imagery, they have widely integrated in autonomous systems to determine the geo-location of objects. The vision-based geo-location systems have found their applications in many aspects, such as environment real time monitoring, search and rescue, reconnaissance missions, and situational awareness. One of the key challenges in the vision-based geo-location systems is to accurately estimate the geo-locations of objects. Since no geo-registration algorithm provides a perfect solution, it is essential that the geo-location systems return accurate estimates of geo-location uncertainties. Accordingly, a precise uncertainty estimation when computer vision algorithms are used in the geo-location process. should be thoroughly studied. UtopiaCompression Corporation (UC), leveraging its unique capabilities and expertise in computer vision, control and estimation, and path planning and mapping, proposes an innovative algorithmic prototype for accurately estimating uncertainty of the vision-based geo-location process. The proposed approach sequentially analyzes the uncertainties for each step of an image-to-map geo-location algorithm and fuses the uncertainties into a final accurate uncertainty estimate of geo-location. A successful feasibility study during Phase I will demonstrate that the proposed algorithm can provide accurate estimates of geo-location uncertainties, thereby corroborating its potential efficacy. BENEFIT: In support of autonomous systems in accurately geo-locating objects using computer vision algorithms, the proposed technology will compute a precise uncertainty estimation for the vision-based geo-location systems. Accurate characterization of geo-location uncertainty will enable deployment of computer vision technique within the military and civilian applications. Within the commercial domain, the key technology areas and related applications that can potentially benefit from the proposed technology include monitoring in shipboard environments using UAVs, civilian search and rescue, reconnaissance missions in contaminated area, and situational awareness of battlefield. All of these applications will benefit from accurate characterization of geo-location uncertainty. UC has identified numerous product opportunities within the US Military modernization effort centering on implementing C4ISR (Command, Control, Computers, Communication, Intelligence, Surveillance, and Reconnaissance) technologies. ISR spending in the next decade is estimated at $15 billion, and depends largely on the stable and accurate navigation of reconnaissance, monitoring, recording and communication devices increasingly unmanned robotic or other UAV devices- which would immediately benefit from UC"s technologies with improved autonomous navigation and cooperation capability, among other benefits.