Video AI

Video AI connects to your camera feeds, such as security, retail, city operations, or industrial cameras, and turns them into structured events you can search, alert on, and analyze. Instead of watching footage, you ask what you want to know.

What Video AI does

Video AI watches your streams the way a very attentive, never-tired security analyst would. It detects people, vehicles, objects, and behaviors; tracks things across frames; raises alerts when something matches a rule you defined; and keeps a searchable log of everything it saw.

You can query the log in plain English. For example:

• "Show me every vehicle that entered the east gate after hours last week."
• "Was anyone in the server room between Friday 6pm and Monday 8am?"

Results come back as a scrollable list of clips with timestamps.

Connect streams

Three ways to bring in video:

• RTSP / ONVIF: most enterprise and municipal cameras.
• Cloud VMS: Milestone, Genetec, Avigilon, Eagle Eye, and similar.
• Upload: for after-the-fact analysis of recorded footage.

Each stream gets a name, a location, and optional metadata about what is in the frame and which area it covers. That metadata is what makes the system useful later. When you ask "anything unusual at the loading bay?" Aleria knows which streams cover the loading bay.

You set how long streams are retained. Events, clips, and full footage can each have their own retention. A common pattern is to keep all events indefinitely, keep clips for 90 days, and not store raw footage at all.

Detectors and rules

A detector is a rule that turns a pattern in the video into an event. The built-in library covers:

• People: count, cluster, presence, loitering, forbidden-zone entry
• Vehicles: count, type, direction, license-plate capture
• Objects: left-behind items, specific categories such as bags, hazards, or equipment
• Behaviors: running, falling, queue length, crowd density, altercations
• Environmental: smoke, fire, water, spills
• Custom: train a detector on your own examples if the off-the-shelf ones do not cover what you need

Build a rule

A rule has three parts:

1. Where: which streams and which regions inside the frame.
2. What: which detector and which parameters, such as "more than 5 people in the queue for over 2 minutes."
3. Then: what happens when the rule fires, such as log an event, send an alert, page on-call, or trigger a workflow.

Dashboards

Every stream can feed a dashboard, which is a Board tuned for video data. Common panels include:

• Live thumbnails of each stream
• Event counts by type over time
• Heatmaps of where activity clusters in the frame
• Dwell-time or queue-length curves
• Alert log, with the most recent at the top

Like any Board, you can ask it questions in plain English and pin the answers as new panels.

Alerts

Alerts fire through the channel you configure: email, SMS, push, Slack, Teams, or webhook. Each alert carries a clip, a timestamp, the detector that fired, and a link back to the live Board. Recipients can acknowledge, escalate, or dismiss directly from the notification.

If the same rule fires a lot, Aleria rolls alerts up into one summary so the channel does not get flooded.

Search footage

Ask the chat directly. For example:

• "Show me every vehicle entering the east gate after 10pm last week."
• "Was anyone in the server room between Friday 6pm and Monday 8am?"
• "Summarize what happened at camera 7 between 14:00 and 14:30."
• "Any incidents involving forklifts in the last 30 days?"

Results come back as a list of clips with timestamps and detector tags. Click any clip to watch it inline.

Video plus workflows

When a rule fires, a workflow can take over: file an incident report, open a ticket, generate a nightly roll-up, or cross-reference with access-control logs. Video AI gives you the event; workflows decide what to do about it.

Privacy and deployment

Video is sensitive, so Video AI is designed so none of it has to leave your environment. It follows the same two access options as the rest of Aleria:

• Aleria Private Cloud: runs in a private, isolated environment in a region you choose, with data residency.
• On-premises: all processing happens in your own data center, so no frames leave the building. Processing can run close to the cameras, and a fully air-gapped install is available for the highest-sensitivity environments.

For more details, see Deployment and security.