ART DeepSense
AI-powered video detection, classification and tracking for surveillance cameras
ART DeepSense is an advanced real-time machine learning module that provides video detection, classification, and tracking capabilities for the cameras of an integrated multi-sensor surveillance solution.
ART DeepSense points automatically the cameras to the threats reported by radars and other sources (AIS, ADSB, RemoteID, RF Sensors…) and uses the latest machine learning algorithms to detect, classify and track objects such as drones, birds, cars, persons, and ships. With the ability to process multiple video streams from the same camera (visible and thermal), ART DeepSense provides a comprehensive solution for surveillance systems. The module allows for the execution of AI models in cascade to classify objects at different levels, providing a fully automatic, flexible, and customizable surveillance solution.
ULTRA LOW LATENCY
Processes video streams with approximately 300 ms end-to-end latency, enabling high speed target tracking.
FAST AI INFERENCE
Runs AI models at up to 25 fps, ensuring smooth real-time detection and classification.
CASCADE AI MODELS
Sequentially executes multiple AI models, refining object classification with each processing stage.
INTELLIGENT TRACKING
Fuses radar and camera data, selecting the most reliable source and adjusting PTZ for continuous tracking.
Highlights
Detect, classify and track
ART DeepSense delivers real-time detection, classification, and tracking of threats reported by radars, AIS, ADS-B, and RF sensors. It automatically points cameras toward detected targets, maintaining continuous visual confirmation, even for fast-moving objects like drones, vehicles, or intruders, ensuring rapid and precise threat assessment.
Enhance any camera
Seamlessly integrate ART DeepSense with any remotely controlled camera, including dual visible and thermal streams. The system automatically adjusts zoom, pan, and tilt based on target distance and location, transforming standard cameras into advanced tracking devices without the need for costly hardware replacements.
Train custom models
Tailor ART DeepSense to your specific operational needs by training your custom AI models for unique threat profiles. Whether detecting hazardous drone types or distinguishing between wildlife and unauthorized vehicles, the platform’s adaptable architecture ensures precise classification for even the most complex scenarios.
Stores forensic evidence
Capture and store synchronized dual-stream video (visible and thermal) up to 4K resolution and 25 fps. ART DeepSense ensures that every critical event is securely recorded, preserving visual evidence essential for post-event analysis, compliance verification, and legal proceedings.
Applications
Technical info
| Functionalities | Multiple detection, classification and tracking of potential threats and swarms |
| AI engine | Real time AI models inference Backend: CUDA, DirectML, TensorRT and CPU Support |
| Visualization | Dual video stream (Visible & Thermal) real time visualization Head Up Display visualization Detections and tracks bounding boxes |
| Video storage | Dual video stream storage (Visible & Thermal) |
| PTZ cameras & optronics | On-screen control Automatic zoom and tilt correction based on distance Pelco-D, Onvif or third-party protocols |
| Cascade AI models execution | Multi-threaded cascade model execution management for accurate identification of sub-classes |
| Custom models | Supports custom-trained models tailored to specific application requirements |
Technologies
Low latency
Real time edge AI execution over selectable video streams
Cascade execution of AI models to provide granular classification
ART R&D team has a broad scope of research and engineering background that results in an extensive range of cutting edge capabilities.
