ReCAP shot lists through actor presence and screen time

Using ReCAP to generate shot lists based on actor presence and screen time turns a long recording into a traceable cast record. Instead of manually noting every appearance, a production team can see who enters a shot, how long the person remains visible and how their presence accumulates across a programme. The output is an editorial working document, not a substitute for human judgement or a speaking-time transcript.

The EU-funded Real-time Content Analysis and Processing project addresses that need through automated video analysis, including metadata extraction, quality monitoring and content recognition. A generated shot list can connect technical detections with editorial questions: who is present, how often they appear and whether a sequence contains material already seen elsewhere in the archive.

Australian production makes this particularly valuable. Stories can connect a Sydney studio with a Melbourne interview or a Brisbane location shoot, while teams prepare material for public broadcasters, commercial networks and streaming services. Timecode, source identity and the delivered edit must remain clear, because the same programme may exist as a full broadcast, a shortened version and several promotional clips.

Before deployment, teams still need clear rules about footage, consent and measurement. Screen time might mean any visible face, a recognisable face or an identity-confirmed appearance. Australian privacy obligations, including the Privacy Act 1988 where applicable, also shape how biometric data and production footage are governed. Definitions should accompany data retention and access decisions, particularly for footage involving children, bystanders or sensitive locations.

From detected appearances to a usable shot list

Start by defining the unit of analysis. A shot may mean a continuous camera view between edits, a detected scene or a segment separated by a gap in recognition. The chosen rule must remain consistent, since counting every camera cut as a new appearance creates misleading totals. For broadcast-quality output, source timecodes should be reconciled with the final edit, especially when separate recordings have been trimmed or rearranged.

For every shot, the useful fields include a shot identifier, source and programme timecodes, actor labels, the beginning and end of each presence, cumulative visible duration and a confidence score. Identity tracking connects the same person across successive shots, while scene or take identifiers preserve editorial context. An editor can then sort by actor, chronological order or programme section without scanning the whole recording.

Several related measures should not be confused. Cumulative screen time totals a person’s visible appearances; longest single appearance measures continuity; average shot duration reveals a different editorial rhythm. A presenter with many short reactions and a guest with one long interview are not equivalent contributors by these metrics. The list records visual presence, not speaking time, importance or approval, and it should not replace the camera report containing lens, framing and movement.

Reading faces and contextual cues

ReCAP’s face detection and recognition provide the core cast signal, but a clean binary label can hide important differences. A small background face and a close-up are both present, yet they offer different levels of prominence. Where useful, face size, position and sharpness can support a visibility score reported separately from identity confidence. This makes screen time an auditable measurement rather than an unexplained total.

Presenter-led footage also benefits from contextual cues. ReCAP’s work on real-time hand gesture recognition illustrates how movements can complement identity matching in interviews, demonstrations and weather segments. A gesture may show that someone is active, but it cannot establish who the person is or what was said. Keeping those signals separate prevents a confident gesture classification from masking uncertain face recognition.

Real footage tests recognition in ordinary ways: changing camera angles, eyeglasses, partial occlusion, crowd backgrounds and abrupt movements. A review interface should let an operator merge duplicate labels, reject a false detection and mark a bystander for exclusion. Corrections should be logged with the original outputs, making the final totals reproducible and showing which names came from an approved cast list rather than automatic labelling.

Separating duplicates from genuine appearances

Archives often contain the same event in several forms. A syndicated segment may appear inside a full episode, an unedited recording and a social clip, while near-duplicate tools may need to tolerate cropping, resizing or recompression. ReCAP’s duplicate detection can expose those relationships, and logo or graphic recognition can add attribution clues. A sponsor card is a visual clue, not proof of ownership or reuse rights.

An unrelated web reference, such as an online casino guide, must be kept outside the broadcast collection rather than processed as a camera-master recording. Otherwise it can pollute catalogue searches, duplicate matching and cast summaries. A sound ingest workflow separates relevant media from unrelated material and flags repetitions for review instead of silently merging everything that looks similar.

Quality problems need their own status. Blur, compression noise, dark frames or severe motion can suppress detection, so the system should expose confidence and missing intervals rather than invent uninterrupted presence. Multi-camera interviews also require a policy: aggregating both camera feeds can double an actor’s apparent time, while showing two actors together means crediting both on that shot. The editorial definition comes before the spreadsheet.

Live production and Australian scheduling

ReCAP’s tools on air offer a practical setting for examining this workflow in a live production. During a rundown, a producer may need to know which presenter is visible, whether a handover has been missed, or how much of the programme a guest has occupied. Real-time metadata can support those checks, while people working on the programme retain responsibility for cues, timing and final editorial choices.

Australian schedules add useful edge cases. Sydney and Melbourne share the Eastern Time zone, while Perth is two hours behind. Daylight-saving changes affect most eastern states, but Queensland does not change, so a Brisbane operation can see clock shifts that differ from Sydney. The list should record the broadcast date and zone for overnight programmes, live crosses and transfers into the archive.

Programme formats also differ across the local market. A bulletin on ABC or SBS does not follow the same commercial-break pattern as material carried by Seven, Network 10 or Foxtel, so the generated list should follow the delivered programme. Sport creates another limit: AFL, cricket, NRL and local club footage can carry separate reuse restrictions. Identifying an athlete’s screen time does not grant permission to reuse the footage.

Exporting results for production decisions

A practical workflow runs from ingest to a reviewable export. The system records source metadata, detects people, tracks identity, associates appearances with shot boundaries and calculates visibility totals. Editors can receive a chronological CSV, structured JSON for a dashboard, or a timecoded document with linked preview frames. Original footage remains the source of truth; the generated list is an index into it.

The resulting answers can be concrete. A team can locate every appearance of one presenter across episodes, check whether a correspondent returns after a break, or compare the duration of studio and location contributions. Filtering repeated copies of an imported segment prevents one busy morning from appearing across several programmes. These uses fit existing production research and archive workflows rather than replacing them.

Every export should carry its assumptions: how shots were divided, which identity labels were approved, what counted as visible, and which frames were rejected by quality checks. Model versions and manual corrections belong with the data so another team can reproduce the result. Clear provenance matters when a numerical total informs commissioning, editorial balance or compliance review.

ReCAP can supply the consistent first pass that manual logging struggles to sustain, especially across long programmes, multiple feeds and repetitive archive material. Its strongest contribution is not a single screen-time figure but a chain of evidence linking an actor’s name to a shot, a timecode and a measurable interval.

The lasting principle is that visibility is measurable while interpretation remains contextual. ReCAP can turn actor presence and screen time into a verifiable production record, but its value depends on definitions, quality controls and human checks that accompany the results.