CULTURAL VENUES · COMPUTER VISION · PLATFORM EXTENSION

Adapting computer vision for gallery security monitoring.

Aurrency monitored an existing IP camera for visitors leaning across a protected boundary and for material changes to an artwork, with configurable regions and timestamped visual evidence.

Client: Aurrency
Sector: Galleries and physical security
Engagement: Two-week proof of concept
Scope: One camera

CAMERA

Existing IP camera

Outbound local connector without public exposure

CONFIGURATION

Configurable regions

Barrier and artwork polygons drawn in-browser

ALERTS

Evidence-first

Timestamp, type and captured image

What the POC established

A two-week, one-camera proof of concept delivered through a configurable platform workflow rather than left as an offline model demonstration.

The walkthrough covered camera selection, pose stages, joint configuration, polygon drawing, schedules and live barrier events.
Client-specific code remained separable from the reusable camera and job platform.

THE PROBLEM AND HYPOTHESIS

A gallery needs to distinguish ordinary visitor movement from the specific conditions that matter.

A rectangular person box does not describe somebody leaning an arm across a barrier. Generic scene change does not distinguish an artwork moving from a visitor briefly standing in front of it.

The installation also needed to adapt to the actual venue. Camera angle, barrier shape, artwork position and opening hours vary. Requiring a source-code change or a newly trained model for each configuration would make a short proof of concept expensive and difficult to evaluate.

Vision platform console for configuring artwork and protected boundary regions on a gallery camera

CONFIGURATION-TO-ALERT WORKFLOW

Configurable rules turned model output into inspectable evidence.

The two-week POC extended an existing platform from one outbound camera connection through configured monitoring to a reviewable alert.

01 · Test stage
Outbound camera connector
02 · Test stage
Scheduled monitoring job
03 · Human decision
User-drawn polygons
04 · Test stage
Pose and change rules
05 · Test stage
Evidence capture
06 · Human decision
Alert review

ENGINEERING DECISIONS

Keeping a two-week camera proof focused and inspectable.

Extend a platform instead of rebuilding the surrounding product

Camera onboarding, job lifecycle, schedules, regions, output storage and live results already existed. The project added a small set of gallery-specific processing stages and retained a clear boundary between reusable Tested Works IP and transferred client code.

Delivery model

Test body joints against the real barrier shape

Selected pose points and arbitrary polygons model a person leaning or reaching more accurately than a rectangular detection intersecting another rectangle.

Computer vision

VALIDATION LIMITS

What the two-week POC did not establish

The work did not establish fleet-scale performance, production security-system certification or a general alarm product. It tested one camera, two defined conditions and an evidence-led workflow on the existing platform.

RESPONSIBILITY