Skip to main content

INVESTIGATE · PERSONS & PATTERN-OF-LIFE

See patterns, not clips.

Every video system gives you clips. Sentinel answers the real investigation question: when is this subject regularly active, where are they anchored, and who moves with them. Open any person and read it on one screen — each part clickable to drill straight into the evidence.

Person profile: 7×24 Pattern-of-Life heatmap with peak cell ringed, co-occurrence network, and location distribution on one screenClick to enlarge

The intelligence picture, on one page. The differentiator no other video platform ships.

7 × 24
Pattern-of-Life
sighting density per weekday and hour, per subject
0–100
Regularity score
predictable vs. random, with a minimum-sample floor
One look
Where they go
anchored, localised, selective, or highly mobile
Click-through
Cell to evidence
any heatmap cell filters the sightings list instantly
The person profile

Everything known about one subject, on one page.

An identity-level record built from face embeddings gives the shape of a subject at a glance — before reading a single chart.

  • Representative face crop — a clear, current face for the subject; set any captured crop as the canonical one with a click
  • Stat grid — total sightings · distinct cameras · captured face crops · first seen · last seen
  • Identity confidence band — an honest read of how well-established this identity is, from Identifying through to Confirmed
  • Lawful-basis card — when the subject is on a watchlist, the basis, document reference, and expiry sit right at the top
  • One-click actions — track on the live map, add to a watchlist, merge or rename — all from the header
Person profile hero: blurred face crop, editable name, identity-confidence band, and the 5-field stat gridClick to enlarge

The confidence band tells an operator whether they are looking at a solid identity or an early one.

Pattern-of-Life heatmap — the differentiator

A week of behavior, in one glance.

One grid reveals the shape of a subject’s week — weekday-morning regular, late-night anchor, weekend-only visitor — without playing a single clip.

  • 168 cells — every weekday-and-hour combination, shaded by sighting density
  • 90-day default lookback, configurable from 1 to 365 days
  • Color scale from quiet to active — dark for no activity, warming through amber to red as density climbs
  • Reads at a glance, drills on demand — the picture is immediate; the evidence is one click away
  • Built for accessibility — the grid carries descriptive labels so it is navigable beyond the visual
Pattern of Life
last 90 days · 339 sightings
MonTueWedThuFriSatSun
0006121823
Peak: Mon 09:00 · 11 sightingsRegularity 78 / 100 · Highly regular

Last 90 days · 339 sightings. Peak cell ringed.

Location distribution

Anchored to one place, or moving across many.

The location distribution reduces a subject's entire movement picture to one verbal band — Anchored, Localised, Selective, or Highly mobile — before the investigator reads a bar.

  • Per-camera bars — each ranked camera with its location label, sighting count, and share of the subject's total
  • A collapsed tail — the top cameras shown individually, the rest rolled into "Other cameras (N)"
  • Anchored — activity is heavily concentrated at a single location
  • Localised — concentrated, but across a small cluster of cameras
  • Selective or Highly mobile — from a handful of places to spread thin across many
Horizontal bar chart: top cameras with location labels, sighting counts, percentage shares, and a collapsed Other cameras row. Concentration band chip reads Highly mobile.Click to enlarge

A subject spread thin across many cameras needs a different strategy from one anchored at one location.

Identity hygiene

One real person, one record — kept clean automatically.

A nightly sweep flags likely-duplicate person records for human review — never a silent merge — keeping every pattern analysis grounded in accurate identities.

  • A nightly sweep compares face embeddings and flags likely-duplicate clusters — erring toward asking rather than auto-merging
  • Cluster-aware — if A matches B and B matches C, all three surface as one cluster to review together
  • A review queue, not silent merges — likely duplicates land at a dedicated queue; nothing is merged without a human decision
  • Side-by-side review — each cluster shows face, name, last-seen, and sighting count for each candidate
  • Merge, keep separate, or defer — and every decision is audited
Persons list with a Likely duplicates KPI tile and amber review banner; the merge queue showing a cluster card with two blurred candidate faces and Merge / Not the same / Skip buttonsClick to enlarge

The nightly sweep proposes. A person decides. Every merge, split, and rejection is audited.

Honest by design

An identity is only as strong as its evidence.

Identity Confidence
14 face samples · 5 cameras
Probable3+ · 0.60–0.75
Honest by design

An operator always knows whether they are looking at a solid identity or an early one — the band never outruns the evidence behind it.

More in the profile

Six analytical lenses, one page.

Regularity score 0–100

The heatmap shows the pattern; the regularity score puts a number on how strong it is — one value from "no clear pattern" to "highly regular," with a minimum-sample floor so a single sighting never masquerades as a confident rhythm.

Peak-cell callout

Above the heatmap, Sentinel names the subject's single busiest weekday-and-hour in plain language — "Peak: Monday at 09:00 (11 sightings)" — and rings that cell on the grid, so the words and the picture are always describing the same thing.

Cell-to-evidence drill-down

Click any cell on the heatmap and the sightings list below filters instantly to that weekday and hour. The clicked cell holds its marker; a clear chip on the list header shows which slot you're reading. Click again to clear. One click from pattern to proof.

Location distribution

Per-camera bars rank where the subject spends their time, with a collapsed tail for high-mobility subjects and a concentration band — Anchored, Localised, Selective, or Highly mobile — that turns the distribution into one plain answer.

Identity confidence band

Every profile carries an honest read of how well-established the identity is: Identifying · Few samples · Low confidence · Probable · Confirmed. An operator always knows whether they're looking at a solid identity or an early one.

Duplicate detection & merge

A nightly sweep proposes likely-duplicate clusters for human review — never a silent merge. Side-by-side review, merge or keep separate, every decision audited. One real person, one record, kept clean without manual effort.

How it works

From a face to a full pattern of life.

1

Identity is built from faces

As cameras run, the AI vision pipeline extracts faces and turns each into a numeric signature. Matching signatures roll up into one person record, so a subject's appearances accumulate behind a single identity rather than scattering across unlinked clips.

2

The profile rolls up the history

Open a person and Sentinel aggregates every sighting into the stat grid, heatmap, regularity score, peak callout, and location distribution — computed against a configurable lookback window, in time to read on the page.

3

Read the shape, then drill the evidence

Take in the pattern at a glance — when they peak, how predictable they are, where they're anchored — then click into any cell or camera to pull the exact sightings behind it, each with a jump to the event and its footage.

4

The identity stays clean and accountable

A nightly sweep proposes duplicate merges for human review; merges, splits, and renames are audited; and where a subject is watchlisted, a lawful basis with an expiry rides at the top of the profile and every query against it is logged.

Under the hood

Specifications

Identity backing storePostgreSQL pgvector (512-dimension ArcFace face embeddings)
Pattern-of-Life grid7 × 24 (weekday × hour), sighting density per cell
Pattern-of-Life lookback1–365 days (default 90)
Regularity score0–100, derived from Shannon entropy across the 168 cells; uniform spread → 0, full concentration → 100
Regularity minimum-sample floorScore held down below the minimum sighting count (default 10) so a single sighting cannot score 100
Regularity bandsNo clear pattern · Some pattern · Moderately regular · Highly regular
Peak-cell calloutBusiest weekday-and-hour, named in plain language + ringed on the grid
Cell-to-evidence drill-downClick any cell → sightings list filters to that weekday + hour; reversible; honest empty state
Location distributionTop-N camera buckets (default 15) + arithmetic "Other cameras (N)" tail
Concentration bandsAnchored · Localised · Selective · Highly mobile
Identity confidence bandsIdentifying · Few samples · Low confidence · Probable · Confirmed
Duplicate detectionNightly per-organisation kNN sweep; union-find clustering; tighter threshold than live detection; human review queue
AccountabilityLawful basis with expiry on watchlisted subjects; every profile action written to a tamper-evident audit

Specifications describe shipped platform capabilities; we'll confirm the configuration that fits your deployment during your demo.

Open a real profile and read its pattern of life.

Request demo access and we’ll send you a private, pre-loaded environment. Open a person of interest, read their 7×24 pattern, click a peak hour straight through to the sightings behind it, and see where they’re anchored — the differentiator, on live data.