Overview#
A vehicle appears on a petrol station camera, then on a motorway ANPR feed three minutes later, then on a city centre CCTV an hour after that. Connecting those appearances manually, across hundreds of hours of footage from dozens of cameras, is the kind of work that used to take investigative teams days. The Media Video Analysis module applies computer vision to this problem, tracking objects and individuals across frames and camera views, detecting behavioural anomalies, and surfacing patterns that matter, without requiring an analyst to watch every second.
This capability supports law enforcement surveillance investigations, body-worn camera analytics, military video intelligence, and any operational context where extracting structured insight from video evidence at scale is a requirement.
Key Features#
- Object detection and tracking across video frames with full trajectory analysis
- Activity and behaviour recognition for event detection in surveillance footage
- Scene analysis with environment classification and change detection over time
- Temporal pattern detection identifying recurring events, routines, and anomalies
- Vehicle detection and tracking with colour, make, and model identification where resolution permits
- Person detection with re-identification across multiple camera views and locations
- Video summarisation generating highlight clips of significant events for efficient review
- Batch video processing for large surveillance evidence collections without manual queuing
- Military video analytics support including metadata extraction from tactical video formats
Use Cases#
- Tracking persons and vehicles across surveillance footage to build movement pattern timelines for investigation
- Detecting suspicious activities in surveillance recordings through behaviour recognition, surfacing events that warrant closer review
- Generating video summaries that condense hours of footage into the significant moments, allowing analysts to cover more ground faster
- Re-identifying the same individual across multiple camera angles and locations to establish presence and movement
- Processing military video intelligence feeds for object detection, activity logging, and tactical event extraction
Integration#
Media Video Analysis connects with evidence management, surveillance system integrations, and investigation analysis workflows. Detection outputs, including object tracks, behaviour events, and re-identification matches, are packaged as structured data alongside the source video and stored in Cloudflare R2 with cryptographic integrity checksums for court-admissible evidence packaging.
Open Standards#
- STANAG 4609 / MISB ST 0601: Tactical video metadata is ingested by parsing Key-Length-Value (KLV) local sets conforming to this NATO/MISB standard, extracting sensor position, platform attitude, slant range, and frame-centre coordinates from UAV and fixed-camera streams.
- SMPTE 336M (KLV Key-Length-Value Encoding): The KLV universal label and BER-OID length encoding defined by SMPTE 336M are used to locate and decode MISB metadata packets embedded in MPEG transport streams.
- ONVIF (Open Network Video Interface Forum): Camera discovery, capability negotiation, and stream URI retrieval for IP surveillance cameras use the ONVIF Device Service and WS-Discovery multicast profiles.
- RTSP (RFC 2326): Real-Time Streaming Protocol is used to connect to and ingest live video feeds from RTSP-capable cameras and relay nodes.
- WGS 84: All geocoded detection coordinates, including sensor latitude/longitude, frame-centre position, and drone telemetry, are expressed in the WGS 84 geodetic datum as decoded from MISB KLV tags.
- SHA-256 (FIPS PUB 180-4): Evidence clips and surveillance recordings are hashed with SHA-256 on ingest and re-verified after storage, providing a cryptographic integrity record for court-admissible evidence packages.
- GraphQL: All detection queries, mutations (ingest, model registration, class management), and real-time detection subscriptions are exposed through a typed GraphQL API.
Last Reviewed: 2026-02-05 Last Updated: 2026-04-14