AI CCTV monitoring
AI that watches every camera, on your own server.
Turns the CCTV you already have into checks for PPE, fire and smoke, restricted zones, people, vehicles and visitors, with a picture and a clip behind every alert. No cloud.
White-label: the licence sets the product name, logo and colours.

Safety checks depend on who happens to be looking
It watches every camera, every shift.
Walk-arounds cover one area at a time.
A supervisor sees what is in front of them, for as long as they are there.
CCTV records and nobody watches.
No team can watch every camera through every shift.
Incidents are found after the fact.
A missing helmet or a person in a restricted area becomes a report, an injury or an audit finding, days later.
One server checks every camera and keeps the proof
Watches
Checks for missing PPE, fire and smoke, people in restricted zones, risky moves near forklifts, counts, plates and visitors.
Any RTSP camera or recorder works.
Proves
Opens an event with a picture and a video clip, and alerts the right people.
Reviewers mark each event a real violation or not a violation.
Learns
Trains on footage from your own site, reviewed by your own team.
A person activates every new model.
From camera to evidence to a better model
Your reviews train the next model
- 1
Cameras
Any RTSP camera or recorder.
- 2
Detect and track people
Finds each person and follows them from frame to frame.
- 3
Run checks per zone
PPE, safety, security and counting checks, switched on per camera and zone.
- 4
Warn, then confirm
A person turns amber first. An event opens only if the item stays missing for the rule's set time.
- 5
Event with evidence
Annotated and raw frames, plus a video clip.
- 6
Review
Your team marks each event a real violation or not a violation.
- 7
Site-trained model
Reviewed frames train a model on your site. You activate it.
Every camera on one wall, every person tagged
Compliance tags.Each person carries one. Missing PPE shows in amber.
Built for a control room.Auto, focus and column layouts, plus a kiosk mode for a wall display.
Live video, boxes on top.One switch shows or hides the boxes.

Every violation arrives with its proof


Frames and a clip.Annotated and raw snapshots, plus a video clip cut from the recording.
One page per event.Missing PPE, severity, camera, zone and person id together.
Review and notify.Reviewers mark each event a real violation or not a violation. Confirmed violations become incidents, with an evidence pack to download. The bell shows high-severity events, incidents assigned to you and system alerts.
Nothing fires on a single frame.
Each rule sets a warning time, when a person turns amber, and a violation time, when the event opens. Each person raises one event per episode.
Rules per zone, with severity and a schedule

Zones
Draw zones on any camera. Each person is judged by the zone they stand in, after a short entry grace period.
Severity
Set it for each PPE type in each zone, with a warning time in amber before each violation.
Schedules
Limit a rule to set weekdays and hours, for requirements that follow the shift pattern.
- Safety helmet
- Safety vest
- Goggles
- Gloves
- Safety boots
- Face mask
- Your own types
Six are on by default. Coverall, ear protection, face shield and safety harness switch on once a model detects them; dust coat and ear plugs are for labelling.
Know where it happens, and try it on your own footage

Reports Available
The compliance summary for any period, filtered by area, zone, camera, PPE type or shift, as a branded PDF or a CSV in your site's time zone. Visits, incidents and vehicles sit beside it as tabs.

Test bench
Drop in an image or a stored video and see what the platform finds, before a single camera goes live. Run it in Fast, Accurate or Max mode, or compare two models side by side.
People across cameras, security when you need it
Follow people across cameras Available today
- One id per person. It follows them from camera to camera, with compliance logged at each sighting.
- Names, groups and merges. Enrol people from photos, name them, set a group or watchlist, and merge two ids that belong to one person.
- Faces are opt-in. Named people are recognised. Under the strictest rules the face of an unnamed person is not kept; a profile can allow longer.
Security mode Available today
- A mode for each camera. A security camera tracks people, logs who was seen where and when, and labels each one with a name, Unknown, or a possible watchlist match.
- Security checks for unknown faces, watchlist matches, authorised-only zones, camera tamper, restricted areas and loitering.
- Available Floor map with your cameras, people shown live and a muster roll-call.
It gets better on your site
Off-the-shelf models miss small PPE on wide, crowded views. Training on your own cameras gives the biggest gain, and the platform has that loop built in.
Two checks before a model goes live
Uncertain frames are picked out for review automatically.
Your team corrects boxes and labels, then approves.
Approved samples become a fixed dataset version.
Runs in idle hours and pauses when the site is busy.
Each candidate is scored on a golden set it never trained on, and compared with the live model item by item.
A person switches it on. Rollback restores the previous model.
No model goes live by itself. Activation needs a passing golden-set score and no item worse than the live model, or a written reason that is kept in the audit log.
Any site with cameras and a safety or security job
Manufacturing and process plants
Helmets, goggles and gloves, set for each machine area. Ear protection once a model detects it.
Construction
Helmets and vests across the site. Safety harness for work at height, once a model detects it.
Warehousing and logistics
Vests and safety boots on the floor, plus person near forklift and person down.
Oil, gas and chemicals
Goggles and gloves in process areas. Face shields and coveralls once a model detects them.
Fire and smoke early warning, to supplement your certified fire alarm.
Mining and quarries
Helmets, vests and boots, plus restricted areas. Ear protection once a model detects it.
Ports and terminals
Helmets and vests across wide yards, plus vehicle counts, number plates and person near forklift.
Utilities and energy
Helmets and gloves, plus zones for authorised people only. Face shields and harnesses once a model detects them.
Food and pharma cleanrooms
Masks, gloves, hairnets and aprons in each cleanroom zone, with a hygiene preset.
ComingDust coat and ear plugs: label now, detect once a model is trained.
After-hours security for any site
Security mode logs who was seen where, and when, on any camera. Security checks raise alerts for unknown faces, watchlist matches, people not allowed in a zone, restricted areas, loitering, and cameras blocked or moved.
Built to run on your site, under your control
One on-site server
Ubuntu, Windows or macOS, with an NVIDIA GPU or Apple Silicon. One-click install on each. Helper servers of yours can share the work.
Works offline
After installation the site needs no internet. A licence file activates it offline, or a product key online. Models and recordings are local files.
Data stays on site
Video, evidence, people data and models live on your server. Privacy rules follow your country, and camera sound is off unless an admin switches it on for one camera.
Roles and an audit log
Four built-in roles and custom roles you define, two-step sign-in. Every change is logged.
HTTPS and encrypted passwords
Traffic runs over HTTPS. Camera passwords are stored encrypted.
The cameras you have
Any RTSP camera or recorder, with presets for Hikvision, Dahua and EZVIZ.
Backups and upgrades
Nightly backups, with a warning when they stop or fail. An administrator installs a signed update file from Settings: one-click updates and rollback.
Remote access, when you want it
Reach the site through an outbound tunnel, with no inbound ports to open.
Fully white-label
Sell it under your own product name, logo and colours, which the licence sets. The sign-in page and every screen use them.
What is live today, and what comes next
Now Available today
- Live wall with compliance tags
- Areas, presets, zone rules, schedules
- Amber warning, then confirmation, per rule
- Evidence, review, incidents, visits
- People ids, opt-in faces, groups, watchlist
- PPE or Security mode per camera
- One event per person; overlap shown as a range
- Overview, Reports with PDF and CSV
- Test bench; left-behind and blocked-exit checks (Experimental)
- CCTV playback, signed exports, PTZ
- Storage forecast, clean-up and retention
- Alarms: siren, email, SMS, WhatsApp, webhook
- Time in compliance, coverage, measured ranges
- Fire and smoke early warning, person down, phone use, forklift, lone worker
- Security checks: unknown face, watchlist, tamper, restricted area, loitering; Object seen rule
- People counting, lines, limits; area analytics, dwell, queues
- Floor map with live people, blind spots and muster roll-call
- Training loop with a validation report
- No new model worse than the live one
- A blind weekly accuracy check per camera
- Approved accounts, two-step sign-in, audit log
- Privacy rules for your country; sound off by default
- White-label set by the licence
- Apps page, video search and incidents with shift handover
- Visitor kiosk with a PPE check, badge and face matching with consent
- Sound alarms and calls for help; time and attendance
- Signed releases; Windows and Mac services
- Vehicle counts, number plates, parking bays
- Conveyor counting: batches, count screen, panel output
- Helper servers with failover
- Ask the site in plain words; All sites page; faces covered before sharing
- Path replay across cameras, with Who was in and evidence packs
Next In progress or next
- Email beyond alarms. Daily and weekly digests, back-to-normal messages and editable wording.
- Online activation. The service is built; it waits for the vendor's activation address and production key. Licence files work offline today.
- Shift handover by email, signed and masked for each recipient.
- Ask the site with a local language model and the site's manuals; today it answers by fixed rules.
- Sinhala and Tamil screens, alerts and reports.
Later Planned
- Plant watch, behaviour and crowd density
- Speech to text and a priority checking budget
One server, every camera app
Sold under your own name. An Apps page switches each one on and off, and the licence decides which a site may use.
CCTV and recording
PPE compliance
Security
Safety checks
Footfall and area analytics
Vehicles and number plates
Video search
Time and attendance
Today's pages Available








How it will look: Accounts Available




How it will look: Client go-live Partly available








How it will look: Sell-ready white label Available


How it will look: Security system Partly available








How it will look: More apps Partly available








Designed screens Partly available








The small details Partly available






A four-week pilot on up to four cameras
Suggested pilotSite survey
Server installed. Up to 4 cameras connected.
Zones, rules and a baseline
Draw zones, set the PPE rules for each area and record a baseline of events.
Review, then train on your footage
Your team reviews events. The model trains on reviewed frames from your own cameras.
Results review
Walk through the results together and agree the next step.
A suggested plan. Timing adjusts to your site.
What we need from you
A camera list with RTSP access
Addresses and logins for up to four cameras, reachable from the server.
A place for the server
A location on the camera network, with a static address.
The PPE rules for each area
Which PPE each zone requires, and how serious a miss is.
A named safety contact
One person who reviews events and joins the results review.
Recommended server
- Operating system
- Ubuntu 24.04 LTS with Docker, or Windows 10 or 11
- GPU
- NVIDIA RTX 4060 Ti 16 GB, RTX 4070 12 GB or better (driver 580+)
- Processor and memory
- 8 cores, 32 GB RAM
- Storage
- 1 TB NVMe plus a recordings drive. Other sizes on the next slide.
- Network
- Internet for the install only. The site then runs offline.
Choose a server for your site
Minimum
Trials, demos and a few cameras
- Processor
- 6 cores
- Memory
- 16 GB
- GPU
- NVIDIA RTX 2060 6 GB or T4 (driver 580+), or an Apple Silicon Mac with 16 GB
- Storage
- 512 GB SSD
- Network
- 1 Gb Ethernet
- Operating system
- Ubuntu, Windows or macOS
Recommended Best value
A typical site, training in idle hours
- Processor
- 8 cores
- Memory
- 32 GB
- GPU
- NVIDIA RTX 4060 Ti 16 GB, RTX 4070 12 GB or better
- Storage
- 1 TB NVMe for the system, models and evidence, plus a recordings drive
- Network
- 1 Gb Ethernet, cameras on their own network
- Operating system
- Ubuntu 24.04 LTS with Docker, or Windows 10 or 11
Best performance
Many cameras, training while live, the thorough Test bench mode
- Processor
- 16 cores or more
- Memory
- 64 GB
- GPU
- NVIDIA RTX 4090 24 GB, RTX 5090 32 GB or RTX 6000 Ada 48 GB
- Storage
- 2 TB NVMe plus a RAID recordings array
- Network
- 2.5 or 10 Gb Ethernet
- Operating system
- Ubuntu 24.04 LTS with Docker
Recordings take about 43 GB per camera per day at 4 Mbit/s, so size the recordings drive for the days you keep. The System page runs a load test on your own server and says how many cameras fit, so the tiers name a use, not a camera count. Example: 4 cameras for 4 weeks at 4 Mbit/s need about 4.8 TB, so a 4 TB drive keeps about 22 days; at 2 Mbit/s (H.265) they need about 2.4 TB, and a 4 TB drive keeps about 45 days.
Start with a pilot on four cameras
Four weeks, one site, and a results review at the end.
- Software
- Version 1.1.834
Code name Sentinel - Code owner
- Altavision
altavision.lk - Author
- Deeghayu Adhikari
deeghayu.netlify.app
© 2026 Altavision. All rights reserved.
