A major U.S. city had 14,847 cameras. After 7 days, the video got deleted - not because they wanted to, but because storing 4.7 petabytes per month at cloud rates would bankrupt the department. Detectives learned to solve cases fast or not at all.
Major U.S. City (Top 10 by Population)
Public Sector - Smart Cities
Edge Video Analytics & Public Safety
Expanso
Pilot in 8 weeks, full rollout in 5 months
9-month payback on infrastructure investment
The city's police commander had a problem. Every week, detectives asked for footage from 8 days ago. Every week, IT had to explain it was already gone. The 7-day retention limit wasn't a policy choice - it was a budget reality. Cloud storage for 4.7PB/month would cost more than the entire IT budget.
We put the analysis near the cameras - at local points of presence like substations and network hubs. Each location runs ML models that index what's happening across nearby cameras - person, vehicle, event type, timestamp. Raw video stays on cheap local storage. Only the index and flagged clips go to the cloud. A 4.7PB problem became a 47GB problem.
Each location runs object detection and event classification for nearby cameras. Instead of storing 4.7PB of raw video in the cloud, we store 47GB of metadata and 230GB of flagged clips. That's a 99.4% reduction.
Detective needs 'red sedan, Tuesday 2-4pm, near intersection X'? Search returns 47 clips in 3 seconds. Old way: watch 48 hours of footage from 6 cameras.
Raw video stays on local NAS at each camera cluster. Cost: $0.003/GB/month vs $0.023 for S3. We went from 7-day retention to 5 years for less money.
The police commander stopped getting complaints about missing footage. Detectives went from 3 days of video scrubbing to 4 hours of targeted review. The 18 operators now focus on alerts instead of watching feeds. And the budget actually went down.
If your retention limits are budget-driven, not policy-driven, we should talk. We've deployed at cities, campuses, and transit systems.