Scaling Video Analytics Systems to Large Camera Deployments
share
New computer vision techniques, which enable accurate extraction of insights from videos, and the deployment of cameras en masse have made many previously inconceivable applications possible. Scaling video analytics to massive camera deployments, however, presents a tremendous challenge, as cost and resource consumption grow proportionally to the number of camera feeds. This paper is driven by a simple question: can we scale video analytics in such a way that cost grows sub-linearly (or even remains constant) as we deploy more cameras, while the accuracy of the analytics remains stable (or even improves)? We believe the answer is yes. Our key insight is that as cameras are densely installed in a physical space, their video feeds become increasingly correlated with each other, correlations that can be harnessed to improve the cost efficiency and accuracy of multi-camera video analytics. By explaining use-cases and technical challenges, we shed light on the potential of leveraging cross-camera correlations to scale out video analytics to many cameras, and call on the community to unlock this potential.
READ FULL TEXT