We release updated versions as we find ways to improve detection and tracking over time.
Our AI models
Videoloft uses AI models to detect and track activity in video surveillance footage. Our models power Smart Motion, People Counting, PPE Detection and Licence Plate Recognition.
We test and improve them using real CCTV footage, benchmark new releases and publish model updates and known limitations.
How we develop and improve our AI analytics models
Test
We test our models using footage from real-world security environments.
Publish
We publish accuracy results where available and explain known limitations that may affect performance.
Improve
How are AI models are used
Smart Motion
Smart Motion detects people and moving vehicles in CCTV footage and tracks where they move within the camera view.
This makes it easier to search footage for people and vehicles, see movement within specific areas, find line crossings and zone entrances/exits and create Smart Motion Alerts.
License Plate Recognition
License Plate Recognition detects vehicles and number plates and picks the clearest image for recognition.
Our vehicle detection model is trained using real CCTV footage and achieved 97% overall accuracy in V16 benchmark testing.
97%
Accuracy benchmark
People Counting
Detects people crossing a user-defined line and records their direction of travel, providing entry, exit and footfall data.
People Counting V1 achieved 92% overall counting accuracy in Videoloft benchmark testing.
92%
Accuracy benchmark
PPE Detection
Checks whether people are wearing hard hats and high-visibility clothing as they enter defined areas.
The model is trained using CCTV footage from construction sites and achieved 91% overall accuracy in our published benchmark.
91%
Accuracy benchmark
How we develop our AI
More than 1,000 cameras have been specifically opted into our analytics training program, giving us access to real-world security footage that most AI developers don’t have.
This footage covers a broad range of environments worldwide, from warehouses, retail stores and offices to outdoor sites. It allows us to train and test our models across different weather conditions, lighting levels, camera angles, distances and levels of activity, as well as challenging situations where people or vehicles are partly obscured.
We test every model update before release and publish model-specific results, known limitations and areas we’re working to improve.
AI for the cameras you already have
Videoloft combines cloud video management with AI analytics for compatible existing CCTV cameras. Add Smart Motion, LPR, People Counting and PPE Detection without replacing your current camera setup.