Construction Safety
PPE and zone rules on-site
Hard-hat, vest, and restricted-zone detection with alerts to the site team.
ExploreAtomVision: Video AI Analytics at the Edge
AtomVision delivers Video AI analytics through the AI Edge Analytic Box — an on-premise appliance that connects to existing IP cameras via ONVIF. Applications include construction safety (PPE and zone detection), shopping mall people counting, traffic monitoring with LPR, and enterprise facial recognition with consent-first controls.
Catalog appliances: Standard (RK3568, up to 4 AI streams, $299) and Professional (RK3588, up to 16 AI streams, $799). Enterprise is on-premise on custom hardware. Inference runs on the edge box; optional cloud dashboard for multi-site viewing and search.
Video AI Analytics · Edge Processing
Video AI on the cameras you already have.
AtomVision’s AI Edge Analytic Box runs construction safety, people counting, traffic monitoring, and facial recognition on existing ONVIF or RTSP cameras. Standard handles up to 4 AI streams; Professional up to 16. Inference stays on-site (RKNN); add the cloud dashboard only if you need multi-site ops.
- Processing
- On-site
- AI suites
- 4
- Cameras
- ONVIF
Edge box · Site 7 · 4 streams · RKNN
Local inference · on-device NPU
Zone · person
Worker · PPE on
CH03 · Crane zone
ABC-4421 · allow
3 alerts · edge buffer OKYOLOv8n · RKNN
Safety
Count
Traffic
FR
What AtomVision is
AtomVision is edge Video AI on the AI Edge Analytic Box. Plug the appliance into your LAN, connect existing ONVIF or RTSP cameras, and enable construction safety, people counting, traffic monitoring, or facial recognition. Inference uses the on-device NPU (RKNN). Standard covers up to 4 AI streams ($299); Professional up to 16 ($799). Add the cloud dashboard only if you need multi-site ops.
- Keep your cameras — ONVIF / RTSP ingest
- Local alerts — no cloud round-trip required
- Four application suites on one edge box
- Optional cloud for multi-site viewing and search
Built for security and operations teams in retail, healthcare, logistics, and property
NR
Northgate RetailCH
Cedar HealthPL
Pacific LogisticsVI
Vance IndustrialHG
Halverson GroupRS
Ridgeline SchoolsBM
Brightside MarketsAP
Anvil PropertyNames above are illustrative sector examples, not a customer list.
Applications
Four suites. One edge box.
Install the AI Edge Analytic Box, then enable the modules that match the site. Each suite is meant for a specific camera layout and alert workflow.
Mall People Counting
Footfall at doorways
Entrance counts, occupancy estimates, and heatmaps for shopping centers. Accuracy depends on camera angle.
Explore
Traffic Monitoring
Lane-level flow & LPR
Vehicle counts, incident flags, and optional gate plate reads processed on the edge box.
Explore
Facial Recognition
Watchlist & access control
Match faces to allow/deny lists with consent-first enterprise deployment.
ExploreVision Engine
Four application suites. One inference pipeline.
AtomVision Vision Engine runs on the AI Edge Analytic Box — RKNN models on the on-device NPU, local alert rules, and optional cloud analytics for search and long-term trends.
Live · Edge box · Multi-app overlay
PPE · no hard-hat
Count · +1 in
LPR · ABC-218
Edge · 18ms
YOLOv8 · RKNN
Safety + traffic + counting active4 streams
Latency
18 ms
Streams
4 / 16
Alerts/hr
12
Construction safety
Edge safetyPPE, restricted zones, perimeter intrusion — alerts before cloud round-trip.
Mall people counting
ByteTrackEntrance footfall, floor occupancy, heatmaps, and capacity alerts.
Traffic monitoring
ANPRLane counts, incidents, and gate LPR on the same edge appliance.
Facial recognition
Enterprise FRWatchlist matching and access control with consent-first deployment.
All models run on the AI Edge Analytic Box NPU (RKNN). Customer video is never used to train shared models.
Hardware
The AI Edge Analytic Box — compute where the cameras are.
Connect the box to your LAN, point it at ONVIF or RTSP cameras, and run Video AI locally. Alerts are decided on the appliance. Cloud sync is optional, for dashboards and search.
Standard
AI Edge Analytic Box Standard
Up to 4 AI streams
Single entrance, small site, pilot deployments
RK3568 · ~1 TOPS NPU
eMMC + microSD buffer
Dual Gigabit Ethernet
HDMI in / out · 12V DC
Professional
AI Edge Analytic Box Professional
Up to 16 AI streams
Malls, job sites, traffic corridors
RK3588 · 6 TOPS NPU
eMMC, microSD, NVMe option
Dual Gigabit Ethernet
HDMI in / dual HDMI out
On-premise deployment. Hardware is specified with your site — not a catalog appliance.
Enterprise
AI Edge Analytic Box Enterprise
Sized with your site
On-premise deployment on customized hardware
Local inference on your hardware
Storage scoped with the deployment
Multi-site when you add cloud sync
Consent-first FR available
Edge pipeline On-device NPU (RKNN)·ONVIF / RTSP ingest·Local alert webhooks·Optional cloud sync
Outcomes
How teams typically apply the four Video AI suites.
ConstructionComposite outcome
How a multi-site contractor might use edge safety AI on four active job sites.
"The useful change is getting a clip when a zone rule fires, instead of asking someone to rewind a recorder after the fact. Results still depend on camera placement and how the site is run."
Illustrative scenario · not a named customer result
Construction safety application PPE
hard-hat and vest checks
per configured camera
Local
alerts on the edge box
no cloud required
16
streams on Professional
one box per site in this example
Also available
Unified cloud viewing across sites
Optional AtomVision cloud dashboard for live walls, investigations, and semantic video search — synced from your edge boxes when you need central ops.
Security & compliance
On-site processing, access controls, and documented privacy options.
- SOC 2Reports on request
- GDPRResidency options
- BiometricConsent workflows
- HIPAABAA on request
- AES-256GCM, at rest
- TLS 1.3In transit
Frequently asked questions
Straight answers about the edge box, streams, and when you need the cloud.
What is AtomVision?
AtomVision is edge Video AI on the AI Edge Analytic Box. The appliance connects to existing ONVIF or RTSP cameras and runs construction safety, mall people counting, traffic monitoring, and facial recognition on-site. An optional cloud dashboard is available for multi-site viewing and search.
Do I keep my existing cameras?
Yes. The edge box ingests ONVIF Profile S/G/T and RTSP streams from brands such as Hikvision, Dahua, Axis, Hanwha, Uniview, and Reolink when ONVIF is enabled. You do not need proprietary AtomVision cameras.
What does “AI streams” mean versus camera count?
An AI stream is a camera feed that is actively running an analytics model. A site can have more cameras than active AI streams. Standard supports up to 4 concurrent AI streams; Professional up to 16. Exact throughput depends on model mix and resolution.
Standard or Professional — which should I buy?
Choose Standard (RK3568, $299) for pilots and small sites with up to 4 AI streams. Choose Professional (RK3588, $799) for malls, job sites, or corridors that need up to 16 streams or multi-model workloads. Enterprise is an on-premise deployment sized with sales on customer-specified hardware.
How fast are alerts, and do I need the cloud?
Alerts are evaluated on the edge box, so they do not wait on a cloud round-trip. Exact latency depends on camera, model, and scene. Use the optional cloud dashboard only when you need multi-site ops, remote search, or centralized review.
How do I talk to AtomVision?
Email [email protected] with a site sketch or camera list. We can walk through which applications and box tier fit your layout.
Next step
Review the edge box against your camera layout.
Send a site sketch or camera list and we can walk through which applications and box tier fit — construction safety, counting, traffic, or access control.
Or review edge box specs for your stream count.