Overview#
A border surveillance operator monitoring dozens of simultaneous camera feeds cannot manually review every frame. The Video Metadata Extraction module addresses this by applying computer vision and machine learning models directly to live and recorded video streams, automatically generating structured, searchable metadata without requiring an analyst to watch each feed. Objects are identified and categorised, text in the scene is read, and moving subjects are tracked, converting raw footage into a time-indexed record of events that can be queried, alerted on, and passed downstream to other analytical modules.
The module processes video at the point of ingest, enriching each frame with structured annotations before the footage reaches storage. This means analysts can search historical recordings by object type, location, or extracted text rather than scrubbing through video manually. The same metadata stream drives real-time alerting, so a recognised vehicle registration, an unattended bag, or a crowd density threshold can trigger an immediate notification to the relevant team.
Key Features#
- Automated Object Detection: Identifies and categorises object types including vehicles, persons, bags, and equipment across multiple simultaneous video feeds in real time, with confidence scores attached to each annotation.
- Optical Character Recognition (OCR): Reads text visible in the scene, covering licence plates, street signs, vehicle markings, and document surfaces, making extracted text immediately searchable and alertable.
- Facial and Biometric Analysis: Extracts facial feature vectors and gait characteristics for comparison against watchlists or for identifying the same individual across multiple cameras, subject to configurable privacy and consent controls.
- Kinematic Tracking: Calculates speed, direction, and trajectory of moving objects frame by frame, producing structured track records suitable for geospatial analysis and route reconstruction.
- Crowd Density Estimation: Estimates the number of individuals within defined zones and raises alerts when thresholds are crossed, supporting event safety and perimeter management.
- Synchronous Metadata Streams: Outputs a metadata stream time-locked to the original video, enabling overlay visualisations, programmatic event triggers, and frame-accurate downstream processing.
- Privacy-Aware Processing: Applies configurable redaction and access controls so that biometric data is only retained or matched when operators hold the appropriate clearance and the processing purpose is lawfully justified.
- Frame-Accurate Search Index: Indexes all extracted annotations so that operators can retrieve the exact video segment matching a query such as a specific vehicle colour, licence plate prefix, or object combination.
Use Cases#
- Border and Perimeter Surveillance: Detects and tracks persons or vehicles crossing defined boundaries, reads vehicle markings, and raises alerts for known or flagged subjects without requiring continuous manual monitoring.
- Traffic and Fleet Monitoring: Counts vehicle types, detects speeding infractions, and reads licence plates to support urban traffic management, fleet compliance, and enforcement operations.
- Event and Venue Security: Monitors crowd density, detects unattended items, and identifies unauthorised vehicles in real time, triggering immediate alerts to security teams before an incident escalates.
- Incident Reconstruction: Allows investigators to search recorded footage by object, licence plate, or time window rather than reviewing hours of video manually, significantly reducing the time needed to build an incident timeline.
- Critical Infrastructure Protection: Monitors access points and restricted zones around airports, ports, and energy facilities, providing a continuous structured log of activity that supports both real-time response and post-incident audit.
Integration#
The Video Metadata Extraction module feeds its structured annotation output into the broader platform so that other capabilities can act on it without duplicating processing. Geospatial mapping consumes the kinematic track records to plot object trajectories on the operational map in real time. Semantic search uses the metadata index to satisfy natural-language queries against video content. Alert and notification workflows subscribe to specific annotation events, such as a matched licence plate or a biometric hit, and route them to the appropriate team or communications channel. The module exposes a stream-based interface so that third-party video management systems and SIEMs can also consume the metadata output without needing access to the raw video.
Open Standards#
- SMPTE ST 0601 (KLV Metadata): The module encodes extracted metadata using Key-Length-Value formatting as defined by SMPTE ST 0601, ensuring interoperability with military, aviation, and UAV video systems that expect this standard.
- MISB EG 0104 (Predator UAV Metadata): Supports ingestion of legacy and current UAV video streams carrying MISB-compliant metadata, aligning extracted annotations with existing airborne ISR workflows.
- ISO/IEC 19794 (Biometric Data Interchange): Facial feature vectors and biometric samples are structured in accordance with the ISO/IEC 19794 series to support cross-system matching and interoperability with national biometric databases.
- ONVIF Profile S: Connects to IP cameras and video encoders using the ONVIF Profile S standard, allowing the module to receive streams from a broad range of commercial and ruggedised surveillance hardware without proprietary drivers.
- RTSP (RFC 2326): Ingests live video over the Real Time Streaming Protocol, enabling low-latency connection to both on-premises cameras and remotely hosted stream sources.
- W3C PROV-O (Provenance Ontology): Metadata annotations carry provenance records structured using PROV-O, documenting which model version produced each annotation and when, supporting audit and evidential chain-of-custody requirements.
- STANAG 4609: Supports the NATO standard for digital motion imagery with embedded metadata, enabling the module to process video received from allied forces and contribute annotated output back into coalition systems.
Availability#
- Enterprise Plan: Included
- Professional Plan: Available as an optional add-on; biometric analysis and watchlist matching require the Enterprise Plan.
Last Reviewed: 2026-05-26