Commerce tries crowdfunding via a challenge to better analyze AI vision feeds

The U.S. Commerce Department is challenging software developers to find a better way to analyze computer vision algorithms, such as for biometric facial recognition, that are presented with common visual impairments of public safety cameras.

The government wants to improve no-reference metrics and computer vision analysis algorithms.

The Public Safety Communications Research division of NIST (part of the Commerce Department) has posted up to $240,000 in cash prizes for challenge winners. Selected entries will predict image quality and perform root cause analysis, adding this information to computer vision feeds.

Not satisfied with getting analysis of overall frame quality from current no-reference metrics, the government wants analysis of and response to specific areas of a camera’s view obscured by, for example, camera shake, dirt, low resolution feeds, weather and spider webs.

To be considered, entrants must submit data sets of image impairments that cause vision applications to fail. Failure equals an algorithm making an unreliable decision.

They also must show “innovative methods to estimate the failure rate of computer vision algorithms” used on the images and video. The methods most desired are those that work with any vision system and those algorithms written for public safety and first responder roles.

Phase 1 winners will be notified October 30; Phase 2 winners will be notified May 19, 2021. Further details including how prize money will be awarded can be found here.

The federal government is also looking into boosting digital ID use within DHS agencies, running a challenge to create a better user interface for its digital wallet.

Article Topics

AI  |  algorithms  |  biometric identification  |  biometrics  |  computer vision  |  facial recognition  |  research and development  |  surveillance cameras