EAGER: Exploratory Research on Deriving Flight Information from Drone Imagery for Safety Compliance

点击翻译按钮获取中文摘要

Recreational drone use is increasing rapidly in the United States. The Federal Aviation Administration (FAA) has established safety regulations such as flying too high, too fast, in restricted areas, etc. but there is no way to detect violations on a large scale. Further, drone users are unaware of or unconcerned about the regulations since they are self-enforced. Drone users upload large amounts of imagery to the Internet including that from flights which violate the safety regulations. This imagery is often the only evidence of the flights and so an interesting research question is whether image analysis can be used to detect violations from the flight imagery alone. The overarching goal of this project is an automated method to identify specific instances of violations in the large amounts of drone imagery available on the Internet. This would provide valuable information regarding the extent to which the regulations are being violated. It could also be used to pursue specific violators. The project will be done in collaboration with the University of California Center of Excellence on Unmanned Aerial Systems (UAS) Safety (http://uassafety.ucmerced.edu/). This Center provides expertise, support, and training for regulatory compliance, risk management, and the safe operation of UAS across the ten campus UC system.Detecting whether a flight is above the 400 ft limit specified by the FAA will serve as a proof-of-concept. A two-step process will first estimate the spatial resolution (i.e., meters per pixel) of the imagery and then use knowledge or estimates of the camera specifications to compute the height. If successful, the proof-of-concept can be extended to other violations such as flying too fast, above crowds, in poor visibility, etc. Estimating the spatial resolution and height of overhead imagery are novel problems, and the proposed approach is novel, challenging and risky. The project stands to make significant gains. There is currently no way to detect violations on a large scale and so this would be the first solution to this increasingly important problem. And, a broad range of drone image analysis problems beyond height estimation would benefit from knowing the spatial resolution. Results, datasets, and other project artifacts will be made available through the project website.