The Automation Company

02 / SPATIAL AI

Spatial AI

Your existing Hikvision and Dahua cameras become sensors. We run detection models at the edge, on your premises, and raise events only when something needs attention — a missing helmet, a short-loaded truck, a skipped process step.

What it actually does

PPE Compliance

Detection models watch defined zones for helmets, safety shoes, gloves, and vests. A violation raises an event with a snapshot and location — to the safety officer's phone, not a control room screen nobody watches. Weekly compliance rates per zone give you an audit trail for factory inspections.

Stack RTSP/ONVIF ingestion · YOLO-family detection · zone rules · event-driven alerting

Loading Bay Dispatch Verification

Cameras over the bay count cartons and pallets as they cross the tailgate, matched against the dispatch note. Mismatches flag before the truck leaves, not when the customer calls. Every dispatch gets a timestamped video clip attached to its invoice number.

Stack Overhead camera feeds · object counting + ByteTrack tracking · dispatch-note matching

Smart CCTV-based SOP Adherence

For steps where sequence matters — torque before paint, cap before test — vision confirms the step happened and in order. Deviations log against the station and shift, so retraining targets the actual gap. This is verification of process, not surveillance of people, and we configure it that way.

Stack Station-level cameras · action/step detection · sequence rules · deviation events

Facial Recognition

Contactless attendance and restricted-area access from existing entry cameras. Consent is collected explicitly, enrollment is voluntary with a card-based alternative, and embeddings never leave your premises. We put up the statutory signage and configure retention limits before the system goes live — see the note below.

Stack Entry-camera feeds · ArcFace-style embeddings · on-prem gallery · access-control events

Architecture

Data flow: Existing CCTV to Edge Inference to Event Engine to Event Store to Alerts & ReviewExisting CCTVHikvision / Dahua · RTSP/ONVIFEdge InferenceJetson Orin Nano classEvent Enginerules · dedupe · zonesEvent StorePostgreSQL + clipsAlerts & Reviewapp · dashboards
Video never leaves the plant. Only events and short evidence clips are stored, with retention you set.

Deployment requirements

The plant provides

Cameras
Existing IP cameras with RTSP access (we audit coverage)
Network
Access to the CCTV VLAN or NVR
Power
230V point in the server/NVR room
Consent process
Employee notice and consent for facial recognition

We install

Inference node
Jetson-class edge box, one per 8–16 streams
Extra cameras
Only where coverage gaps exist, quoted separately
Signage
Statutory CCTV and biometric notice boards
Review console
Web app for events, clips, and reports

Timeline Pilot on 4 streams in 2 weeks; full plant rollout in 4–8 weeks depending on camera count.

The questions plant heads ask.

Yes, with conditions. Under the DPDP Act 2023, biometric data needs informed consent, a stated purpose, and limited retention. We build all three in: explicit enrollment, a non-biometric alternative for anyone who declines, signage at entrances, and deletion when an employee leaves. Your legal counsel reviews the notice text before go-live.

Ask us about spatial ai.

A short note about your plant is enough — we reply within one working day, or call +91 95990 94602.

Interested in