Computer vision works best with high-quality images. But for public safety officers in the field, things look differently. Computer vision can provide a clear picture of the situation, but dirt and grime, rain, light flares, or simply low-quality camera equipment can disrupt computer vision and remove that valuable tool.
The proposed solution is an algorithm that would go between the camera and the computer vision algorithm. This algorithm would assess the image or video quality. The computer vision system would use this information to deploy complex mitigation strategies, such as zooming, panning, changing the bit-rate, or pre-processing to de-noise, de-block, or remove airborne dust. The system could also make internal changes, such as choosing among several computer vision algorithm.
The Enhancing Computer Vision for Public Safety Challenge is designed to help develop this new line of research. In this challenge, PSCR anticipates solvers will create datasets with camera impairments and an innovative method to assess computer vision failure rates for those media.
NIST held an information webinar on September 10 where the challenge team presented a summary of the challenge and answered participant questions. A recording of the webinar is available above.