SINTRA Multi-Scenario Video Analytics Suite for Airports
- Project
- 22006 SINTRA
- Type
- New product
- Description
TAV Technologies developed and benchmarked a real-time, multi-scenario AI video analytics suite tailored for critical airport infrastructure under SINTRA UC1. The suite comprises five specialized computer vision detection modules—Fire Detection, Fall Detection, Fight/Violence Detection, Left Luggage and Asset/Fixture Detection—designed to pre-screen high-volume security camera streams, suppress false-positive alarms, and minimize event-to-alert latency in terminal operations. Rather than running heavy deep learning inference continuously across hundreds of cameras, the suite supports event-driven verification triggered by acoustic or environmental anomalies and operates over a scalable microservice ingestion and scheduling architecture. Validated outputs are streamed directly to the TAV Technologies AOCC (Airport Operation Control Center) platform to provide human-in-the-loop decision support for security operators without storing biometric or personally identifiable information.
- Contact
- Selim Sarı
- selim.sari@tav.aero
- Research area(s)
- Artificial Intelligence / Data analytics, Safety & Security, Smart mobility / Transport
- Technical features
Multi-scenario AI detection: Dedicated deep learning pipelines benchmarked for four high-priority airport safety and security events: Fire/Smoke Detection, Passenger Fall Detection, Fight/Physical Altercation Detection, and Terminal Asset/Fixture Monitoring. Stream ingestion & orchestration: Microservice-based RTSP video ingestion supported by a dynamic camera discovery, pipeline orchestration, and resource management Scheduler Service. Distributed data backbone: High-throughput event streaming and state management built on Kafka/RabbitMQ message brokers alongside PostgreSQL and Redis data layers. False-alarm suppression: Multi-window temporal persistence checks, confidence calibration, and cross-modal triggering (e.g., thermal/temperature anomalies triggering visual fire verification; acoustic anomalies triggering visual event classification) to reduce GPU load and operator alarm fatigue. Privacy-by-design compliance: Strictly non-biometric processing (no facial recognition or identity profiling) aligned with GDPR and KVKK data minimization principles.
- Integration constraints
Compute environment: Containerized Linux microservice deployment (Docker) requiring NVIDIA GPU acceleration for real-time multi-stream inference. Video input interface: Standard RTSP streams from existing IP cameras or enterprise Video Management Systems (VMS); detection accuracy depends on camera mounting angle, resolution, and severe occlusion levels in dense crowds. Output interface: Structured JSON event payloads published via distributed message queues (Kafka / RabbitMQ / Pub-Sub) or REST APIs to central incident management consoles. Network & security: Requires isolated local VLAN access to CCTV streams; raw video frames are processed in memory and not retained permanently outside existing VMS policies.
- Targeted customer(s)
Airport operators and terminal management companies (AOCC and Airport Security Operations Centers). Other high-density transport hubs: railway stations, metro networks, maritime ports, and intercity terminals. Enterprise physical security and VMS integrators managing critical infrastructure facilities.
- Conditions for reuse
Commercial license for TAV Technologies (available as an add-on module for TAV AOCC or via enterprise API integration).
- Confidentiality
- Public
- Publication date
- 10-11-2026
- Involved partners
- TAV Technologies (TUR)
- ARD GROUP (TUR)
- Necdet Alpata Pazarlama Lojistik ve Turizm Sanayi ve Ticaret A.Ş. (TUR)