3.13.1 Video Analytics Package Update with Enhanced AI Features Released
AxxonSoft has launched an updated version of its video analytics suite for Axxon One. The Axxon DetectorPack 3.13.1 includes new detection tools, the Train It Yourself neural network training service, as well as expanded capabilities and optimized performance for existing tools.
Meta-Detector
The Meta-Detector enables real-time detection of objects or events specified through natural language text queries in English. The optimal query structure includes the object, its key attributes (size, color, etc.), the surrounding environment, and additional scene-specific details, such as angle and object positioning. Non-essential parts of the query can be omitted. Up to three queries can be set simultaneously for a single detector.
Examples of queries:
- Human and dog
- Running human
- Fallen human
- Woman wearing a black dress
- White car pedestrian crossing top view
The Meta-Detector is powered by a neural network capable of recognizing diverse scenarios without additional training. This makes it a versatile detection tool easily configured using natural language.
Neural Network Training Service
The TIY (Train It Yourself) service allows users to train a neural network to identify specific objects. Training can be conducted using a set of static images or webcam footage. The trained neural network can then be uploaded to Axxon One and used with the Object Presence Detection tool, which recognizes the presence of specified object types.
This feature enables our clients and partners to customize video analytics to meet site-specific requirements and surveillance conditions without requiring assistance from AxxonSoft.
Fight Detector VL
The Fight Detector uses artificial intelligence to identify fights based on visual cues such as raised arms or legs, active limb movements, or prolonged proximity between individuals.
VI Scene Analytics
High-quality detectors suitable for use in transportation facilities and other places with mass occupancy have been added:
- Movement in prohibited direction.
- Motion in area.
- Line crossing.
- Atypical changes in the scene (lens obstruction, overexposure, defocusing, camera displacement, or background changes).
Enhancements, Improvements, and Optimization
- Stopped Object Detector: Improved performance on NVIDIA GPUs.
- License Plate Recognition IV: Now supports the NVIDIA Ada Lovelace GPU microarchitecture.
- Handrail Holding Detector: Enhanced accuracy eliminates false alarms caused by brief occlusion of a hand holding a rail, which may occur due to viewing angles or obstructions.
- Track Masking Sub-Detector: Introduced for use with various detector types (e.g., trackers, pose detectors, vehicle detectors) to hide objects during real-time monitoring and video footage reviews.
- Detection Zones: Added settings for defining detection and skip areas across multiple detectors.
- Country-Specific Expansion: Extended the list of supported countries for several license plate recognition and vehicle recognition detectors.
Axxon DetectorPack 3.13.1 is compatible with Axxon One version 2.0 or later. A complete list of changes can be found in the release notes. The basic software package and addons are available on the download page.