Product overview

A vision engine that turns recorded video into real-time understanding

Astra23 watches every frame from your existing cameras and recognizes people, PPE, zones, proximity, and behavior — then matches what it sees against the rules that matter for your context.

LIVECAM-01 · DETECTION ENGINE

--:--:--

PPE: HELMET ✕ — FLAGGED

PPE: OK ✓

1 VIOLATION

ASTRA23 VISION ENGINE30 FPS · 1080p

What it recognizes

Trained on the things that cause incidents

People

Detects and tracks people across the frame, distinguishing known vs. unknown where configured.

PPE

Helmets, high-visibility vests, and harnesses — flagged per-zone against your rules.

Zones

Restricted, active-machine, and no-entry areas defined visually on the feed.

Proximity

Unsafe closeness between a person and machinery or a hazard boundary.

Behavior patterns

Loitering, perimeter breaches, and package interactions over time.

Rule matching

Every detection is evaluated against the active rule-pack for that camera.

Engineering transparency

Honest about operating conditions

Computer vision depends on what the camera can actually see. Poor lighting, extreme angles, and heavy occlusion reduce confidence — so Astra23 surfaces detection confidence, degrades gracefully to a "caution" state rather than a false "clear," and lets you tune sensitivity per camera. We'd rather tell you a camera is struggling than pretend it isn't.

Architecture

A clear path from frame to alert

Camera feed

RTSP / ONVIF / bridge

Vision engine

Edge or cloud

Rule-pack matching

Context rules

Alert + Dashboard

Act & report

How it works in detail Alerts & reporting

Point it at a real feed and watch

The fastest way to understand the engine is to see it run.

Book a live demo Explore verticals