Always on
Every camera is a continuous stream. Analytics that work in a demo on one feed have to hold the same behaviour across hundreds of simultaneous streams, indefinitely, on hardware the customer already owns.
Video Surveillance That Understands What It Is Looking At
AxxonSoft builds intelligent video management software, PSIM and cloud surveillance used to protect sites in more than 140 countries. The engineering problem is continuous: every camera produces a stream that never stops, and the analytics have to keep up with all of them at once.
Build Your Computer Vision PlatformAbout the Project
AxxonSoft is a software company specialising in video management software (VMS), physical security information management (PSIM) and cloud video surveillance (VSaaS), with AI video analytics designed in house. Its Axxon One platform scales from a single small installation to multi-location deployments, and the company states that more than 20,000 customers across 140+ countries rely on its software.
Industry
Physical Security and Computer Vision
Video surveillance and physical security software, where the system is expected to run unattended and be right at the moment something actually happens.
The Platform
Axxon One — video management that scales from one site to many
AxxonSoft describes Axxon One as limitlessly scalable VMS suited to both small installations and large, multi-location deployments, with forensic search and remote monitoring through its cloud service.
AI Analytics
Detection models running against live video, designed in house
The published analytics range includes human and vehicle detection, fire and smoke detection, behaviour analytics, licence plate recognition and object counting, with on-demand training for task-specific cases such as PPE detection.
Scale
More than 20,000 customers across 140+ countries
AxxonSoft's own published company scale. Based in Cork, Ireland, the company operates 32 offices worldwide to support partners and clients across those regions.
Engineering Stack
C++ and Python with OpenCV and TensorFlow, shipped in Docker
The stack recorded for this project: native code where per-frame performance decides how many streams a server can carry, Python and machine learning frameworks for the analytics layer.
Build your idea
Talk to our experts
Scope a computer vision, video analytics or real-time detection platform for your operating environment.
A Camera Does Not Stop, and Neither Does the Cost of Watching It
Video surveillance software is judged on the hours nothing happens as much as on the minute something does.
False alarms
An operator who is alerted too often stops reading alerts. Detection quality in the field is measured less by what a model catches than by how rarely it cries wolf over a shadow, a bird or a headlight.
After the fact
Most footage is reviewed, not watched live. Finding the relevant thirty seconds inside weeks of recording is a search problem, and it is the one investigators actually spend their time on.
Hundreds of live streams,
analysed continuously,
on hardware already in the rack
Build Your Vision Platform
How the Platform Works
-
Ingest and Manage Streams
Ingest and Manage Streams
- Video management is the base layer: cameras, recording, retention and access across one site or many
- AxxonSoft positions Axxon One as scaling from small installations to multi-location deployments on the same platform
- C++ carries the performance-critical path, where per-frame cost sets how many streams a single server can carry
-
Run Detection on Live Video
Run Detection on Live Video
- Published analytics cover human and vehicle detection, fire and smoke, behaviour, licence plates and object counting
- OpenCV and TensorFlow sit behind the analytics layer, with Python as the development surface for model work
- Analytics run continuously rather than on request, which makes throughput per stream the governing constraint
-
Search the Archive
Search the Archive
- Forensic search lets an investigator query recorded video by what was detected rather than by scrubbing a timeline
- Detection output has to be indexed as it is produced, because recomputing analytics over an archive is prohibitively expensive
- This is where analytics quality pays off twice: once as a live alert, once as a searchable record
-
Deploy Across Sites and Clouds
Deploy Across Sites and Clouds
- Docker packaging supports consistent deployment across the varied server environments customers and integrators run
- A cloud service extends the same platform to remote monitoring, where bandwidth rather than compute becomes the limit
- Multi-site installations need one operator view over infrastructure that is physically and administratively separate
What This Build Demonstrates
Computer vision engineered for continuous operation, not for a demo.
Native Performance Where It Matters
C++ on the frame path with Python and TensorFlow for the model layer is a deliberate split: throughput is bought in native code, iteration speed is kept in the machine learning stack.
Analytics Judged in the Field
Detection is evaluated on false alarms in real conditions rather than benchmark scores, because an operator's trust is the metric that determines whether the feature survives.
Built for Integrator Delivery
Software distributed through partners into installations across many countries has to be packaged, upgradeable and diagnosable by people the vendor never meets.
Live and Forensic in One System
The same detection output has to raise an alert now and remain searchable months later, which makes indexing at ingest a core architectural requirement.
More Case Studies
Comviva: Platforms Behind Telco and Bank Services
FinTech, MarTech and DigiTech platforms deployed by operators and financial institutions across more than 100 countries.
PokerBaazi: Real-Time Multiplayer at Scale
Thousands of simultaneous tables kept in sync over mobile networks, with the server as the only authority on the hand.
Global presence
