How ReCAP Detects and Logs Broadcast Video with Incorrect Colour Bars

Colour bars have sat at the heart of broadcast engineering since the analogue era, long before Australian living rooms received their first digital set-top box. In Sydney's Harris Street facilities and the smaller regional hubs in Adelaide or Perth, the SMPTE bar pattern still rolls across a confidence monitor a few seconds before a programme goes to air. The familiar stripes give camera shaders, vision mixers, and quality engineers a known reference for luminance, chroma, and hue. When the pattern is wrong, missing, or replaced with something else entirely, downstream monitoring equipment has nothing reliable to lock onto.

ReCAP has built a dedicated analysis layer that watches every frame of a broadcast feed and flags the moments when those reference bars deviate from expectation. The system does not just blink a warning light on a single scope; it creates a persistent log entry that engineers can query, export, and attach to compliance reports. For an industry that lives or dies by transmission standards, that forensic trail has real practical value.

The work sits inside a broader European-funded research initiative that develops tools for automated metadata extraction, quality monitoring, face recognition, logo detection, and duplicate-content identification. The full scope of the project, including its consortium members and demonstration videos, is documented on the ReCAP project site. Pilot deployments have already run with broadcasters across Europe and the Asia-Pacific, with Australian partners contributing test material from live news, sport, and studio workflows.

What Colour Bars Do in a Broadcast Workflow

The original SMPTE colour bar pattern was designed as a calibration tool, not a piece of programme content. Each vertical strip represents a specific primary or secondary colour at a known brightness, allowing an engineer to set up a camera, adjust a vectorscope, and verify that a transmission chain is passing signal without crushing the blacks or clipping the highlights. The pattern survived the transition from PAL to SDI, and now appears in high-definition and 4K test generators that sit in every Australian control room.

In day-to-day operations, colour bars are usually aired for a short period before a programme begins, or held on a black-and-burst signal during a commercial break when local advertising has not yet loaded. Some networks also transmit a brief burst of bars after a scheduled event so that downstream affiliate stations can rejoin cleanly. The pattern has become so embedded in the workflow that operators often refer to it simply as "bars" or "the test pattern," depending on which generation of engineer they trained under.

When the wrong pattern appears, the consequences range from cosmetic to operational. A subtly miscalibrated bar generator might pass unnoticed for hours, slowly skewing the white balance of every camera in the chain. A more obvious error, such as a still frame from a previous programme frozen on the transmission output, will trigger immediate viewer complaints and can pull engineering staff away from a live event. ReCAP treats both ends of that spectrum as logged incidents, because the cost of investigating after the fact is far higher than catching the issue in real time.

How ReCAP Detects Anomalous Colour Bar Patterns

The detection engine inside ReCAP does not rely on a simple template match. Broadcast feeds vary in resolution, aspect ratio, and colour space, and a pattern that is technically correct on a standard-definition SDI feed may look slightly different when up-converted for a high-definition simulcast. The project team trained the detector on a wide corpus of bar patterns, including the original SMPTE EG 1-1990 specification, the high-definition variant, and the 75 percent colour bars used by some Australian networks for legacy compatibility.

The core algorithm analyses the top, middle, and bottom thirds of the frame separately, because colour bars in a real transmission rarely fill the entire raster cleanly. Embedded captions, station logos, and timecode burns sit on top of the pattern in most modern workflows. By segmenting the frame, the detector can confirm that the underlying structure matches a known bar layout even when overlay graphics obscure part of the image. A confidence score is calculated for each candidate match, and only events that exceed a configurable threshold are escalated into the logging pipeline.

False positives were a significant concern during the early pilot phase. Some networks use modified bar patterns during equipment commissioning, and a generic detector would flag those legitimate engineering windows as errors. ReCAP now supports a station-specific whitelist that allows known variants to pass without raising an alert, while still catching genuine misconfigurations. The whitelist can be updated through the same dashboard that engineers use to review flagged incidents, which keeps the configuration close to the workflow rather than buried in a separate tool.

Logging Incidents for Forensic Review

Each detection event triggers a structured log record that captures the technical context of the anomaly. The record includes the precise time the pattern appeared, the duration it remained on air, and the source identifier of the feed where the issue was observed. For multi-channel operations, the system also captures which programme stream was affected and whether the anomaly coincided with a scheduled event such as a commercial break, a sports cross, or an emergency bulletin.

The metadata fields stored per incident include:

These records are written to a searchable database that engineering teams can query by date, channel, or operator. For Australian broadcasters that need to demonstrate compliance to ACMA or to internal quality committees, the ability to pull a complete incident report from a single dashboard removes the manual collation work that used to follow a problematic transmission. Logs can be exported as CSV, JSON, or PDF, depending on the downstream system that consumes them. For engineers who want to dig into the algorithm choices and the trade-offs made during development, the team has posted a detailed write-up on the project blog, which covers the same ground in more depth.

Integration with Live Production and Asset Management

ReCAP is designed to sit alongside existing broadcast infrastructure rather than replace it. The detection module runs as a service that consumes live feeds through standard SMPTE ST 2110 or compressed transport streams, depending on the facility. For file-based workflows, the same engine can analyse mezzanine files after they have been ingested into a media asset management system, flagging any pre-recorded content that carries an incorrect bar pattern at its head.

Integration with alerting platforms follows common broadcast conventions. When the detector identifies a confidence score above the configured threshold, it can fire an SNMP trap, post a message to a Slack or Microsoft Teams channel, or update a tally light on a control room wall. Australian operators who manage regional studios from a central hub in Sydney often prefer the messaging channel approach, because it allows junior engineers on shift to receive notifications without leaving their main monitoring application. Larger networks tend to wire the alerts directly into their existing incident management systems, where they appear alongside equipment faults and staffing issues.

The logging output is deliberately compatible with the EBUCore metadata standard, which most Australian broadcasters already use inside their MAM platforms. That compatibility means a flagged incident can carry through to archival systems, so that even years later, an engineer investigating a viewer complaint can reconstruct what was on air and whether any technical anomalies coincided with the report. The forensic value of that continuity has proven especially useful during long-running investigations involving advertising standards or content classification disputes.

Australian Standards and Operator Workflows

Free TV Australia publishes the operational guidelines that most commercial networks follow, and ACMA enforces the underlying technical framework that requires broadcasters to maintain signal quality. While the regulations do not specifically mandate the use of colour bars, they do require that transmission equipment operates within defined parameters and that any deviation is logged for review. ReCAP's automated logging provides the kind of continuous record that makes those reviews straightforward, particularly during high-profile live events where multiple feeds converge in a master control room.

Outside broadcast operations in Australia present their own challenges. Trucks covering the Sydney New Year's Eve fireworks, the Australian Open in Melbourne, or a State of Origin match in Brisbane often work with mixed feeds from host broadcasters, satellite uplinks, and on-site cameras. A bar pattern that looks correct at the source can arrive at the network head-end with timing offsets, embedded captions, or regional graphic bugs that confuse a simpler detector. ReCAP's frame-segmentation approach was tested against these scenarios, and the project consortium has documented specific configuration profiles that Australian operators can apply without retraining the underlying model.

Smaller regional broadcasters have found the tool useful in a different way. Stations that do not run a 24-hour engineering shift often rely on overnight transmission automation, and a misconfigured bar pattern that persists for several hours can go unnoticed until morning. ReCAP's persistent logging surfaces those overnight incidents in a morning digest, giving regional engineering teams a clear starting point when they begin their day. The digest integrates with the same notification channels used for live alerts, so nothing is hidden in a separate reporting tool.

Pilot Outcomes and Practical Limits

The pilot deployments have produced measurable results across both European and Asia-Pacific sites. Detection accuracy on standard SMPTE patterns sits above 99 percent in controlled tests, with false positive rates dropping below 0.2 percent once the station-specific whitelist is configured. In live operation, the system has caught genuine misconfigurations within seconds, allowing operators to correct the issue before it propagated to a wider audience.

The team has also documented the typical triggers that prompt a colour bar alert in real facilities:

These categories cover the majority of incidents observed during the pilot period. Some edge cases remain, particularly when a broadcaster uses a heavily branded custom pattern that deviates significantly from the SMPTE specification. In those cases, ReCAP's configuration layer allows operators to define the expected pattern manually, ensuring that legitimate variations are not flagged as errors. The system is not designed to replace human judgement during complex live events, but it does provide a reliable safety net that operates continuously, even when the control room is fully occupied with a high-stakes broadcast.

Engineers interested in testing the detection module against their own feeds can request access through the consortium's demonstration programme, which provides a sandboxed environment with sample Australian content already loaded and ready for analysis.