Video Management
VMS
A complete video management system on the box: live multi-channel view, continuous and event recording, timeline playback and export — no separate NVR or server.
Solutions
A single edge device that detects, manages and understands video — from raw camera streams to searchable, structured intelligence.
Detection engine
Person Attributes
Understand who is in frame, not just that someone is. The box extracts structured attributes for every person, continuously.
- Age range & gender estimation
- Upper / lower clothing colour
- Bags, backpacks & accessories
- Headwear & mask presence
- Re-identification across channels
PERSON ATTRIBUTESCH-01
Age range & gender estimation0.950
Upper / lower clothing colour0.941
Bags, backpacks & accessories0.932
Headwear & mask presence0.923
Detection engine
Vehicle Recognition
Turn every lane and gate into structured data. Classify, describe and read vehicles for traffic, parking and access control.
- Type: car, van, truck, bus, motorcycle
- Colour & make classification
- Licence-plate recognition (LPR)
- Direction & speed estimation
- Cross-channel vehicle tracking
VEHICLE RECOGNITIONCH-02
Type: car, van, truck, bus, motorcycle0.950
Colour & make classification0.941
Licence-plate recognition (LPR)0.932
Direction & speed estimation0.923
Detection engine
Action Detection
Catch behaviour the moment it happens. Rule-based and learned actions raise alerts without a human watching every feed.
- Walking, running & falling
- Loitering & crowd forming
- Line crossing & intrusion zones
- Abandoned-object detection
- Custom actions via MLOps
ACTION DETECTIONCH-03
Walking, running & falling0.950
Loitering & crowd forming0.941
Line crossing & intrusion zones0.932
Abandoned-object detection0.923
The platform
Detection is only the start. OsonVision manages your video, evolves your models, and lets anyone search footage in plain language.
Model lifecycle
MLOps
Capture edge cases, label them, retrain and deploy new models straight to the box. Version, roll back and improve accuracy on your own footage — on-premise.
Vision-Language
VLM
A Vision-Language model lets anyone search video in plain language and get scene summaries — "person in red jacket near the entrance" returns the exact clips.
Where it runs
Built for real-world deployments
Retail & Malls
Footfall, demographics and dwell time, plus queue and shoplifting alerts — without cloud subscriptions.
Traffic & Smart City
Vehicle counting, classification and LPR for junctions, toll lanes and restricted-access streets.
Security & Access
Intrusion, loitering and line-crossing alerts with person re-identification across every camera.
Industrial Safety
PPE and helmet checks, restricted-zone entry and fall detection for factories and construction.
Match the AI Box to your use case
Tell us your cameras and goals — we will show the exact detection, VMS and VLM setup for it.