Open-source real-time hoistway awareness for technician safety

Most serious incidents are not mysterious failures. They are moments of uncertainty: unseen movement, unexpected state changes, bad line of sight, or wrong assumptions at the worst possible time.

HoistwayWatch is an open-source project to provide 📹 continuous visibility 📹 and real-time alerts for elevator hoistway work using low-cost cameras plus simple, explainable detection.

This is not meant to replace existing safety practices. It adds real-time awareness so fewer hazards become surprises.

• Real-time hoistway awareness with inexpensive hardware
• Detect motion and unsafe conditions and alert immediately
• Keep logic auditable, simple, and explainable
• Run locally on-site (privacy-first)
• Work vendor-neutral and avoid proprietary controller integration for the MVP

• This is not a controller override system
• This is not a substitute for proper procedures, training, or codes
• This is not a remote monitoring service by default
• This project does not provide OEM-specific instructions

A camera watches the hoistway zones that matter. The system detects:

• Car movement (or any movement in defined zones)
• Human presence in defined danger zones (optional)
• Door zone changes (optional)
• Unusual motion patterns (optional)
• “Movement while someone is present in the hoistway” style rules

Then it triggers:

• Local audible alarm
• Flashing strobe
• Phone notification (local network)
• Event log for review

• One or more cameras with a simple live dashboard
• Timestamped recording optional

• Define regions of interest (ROI) for “car path”, “counterweight path”, “pit”, “overhead”
• Trigger alert when motion occurs in a zone

• Detect a person in a defined zone
• Trigger stronger alert if motion happens while presence is detected

• On-device alarms
• Local network notifications
• No cloud required

• Rules are simple JSON or YAML
• Every alert explains exactly what triggered it

HoistwayWatch is an awareness layer, not a safety interlock.
It should be used only as an additional aid with proper training and procedures.

If you want a safety interlock system, that requires a totally different engineering and certification path.

• Raspberry Pi 5 or Pi 4
• 1 to 2 USB cameras or PoE IP camera
• Small powered speaker + strobe
• Optional local Wi-Fi router (jobsite isolated network)

• NVIDIA Jetson Orin Nano or Orin NX
• 1 to 4 PoE IP cameras
• PoE switch
• Speaker + strobe
• Rugged case

• Simple vibration sensor
• Magnetic door switch for access points
• IR beam for doorway presence

• hw-capture
  Camera ingest (RTSP for IP cams, UVC for USB cams)

• hw-vision
  Motion detection and optional object detection

• hw-rules
  A small rules engine that evaluates events like:

  • motion_in_zone
  • person_in_zone
  • motion_and_person
  • camera_offline
  • low_light

• hw-alerts
  Audible, strobe, push notifications (local), event log

• hw-ui
  Local dashboard showing live feeds, zones, and alert history

Start with simple, robust primitives:

• Motion detection (frame differencing, background subtraction)
• Zone-based triggering
• Optional person detector (YOLO or MobileNet) running locally

Design rule: If the ML fails, the system still provides continuous visibility.

hoistwaywatch/
  README.md
  docs/
    safety.md
    hardware.md
    installation.md
    field-setup.md
    faq.md
  apps/
    dashboard/         # web UI
  services/
    capture/           # camera ingest
    vision/            # motion + detection
    rules/             # event rules engine
    alerts/            # audio/strobe/notify
  configs/
    example-site.yaml
    example-zones.json
  scripts/
    setup-pi.sh
    setup-jetson.sh
  licenses/

• Start the capture service with a webcam
• Open the dashboard
• Draw zones
• Trigger motion and confirm alert

• Use an isolated Wi-Fi router
• Place camera where it can see the critical path
• Add a speaker and strobe
• Validate “movement equals alert” works every time

The system is only as good as camera placement. The MVP guidance:

• Favor angles that show the car path clearly
• Avoid direct glare and strong reflections
• Add a second camera for pits and overheads when possible
• Prefer fixed mounts and stable reference points
• The goal is not pretty video. The goal is clarity under real conditions

• Default mode is local-only
• Recording is optional and off by default
• No cloud upload by default
• If an org wants centralized logging, it’s an opt-in module

• Live view dashboard
• Zones + motion detection
• Local alerts
• Event logging

• Optional person detection
• Multi-camera support
• Better low-light handling

• Portable “jobsite kit” reference build
• Better calibration tools
• Offline-first mobile companion view

• Integrations that are vendor-neutral (not controller-specific) where feasible
• Formal validation guidance for organizations that want to adopt responsibly

We want:

• Field feedback on what zones matter
• Camera placement guidance from actual installers and service techs
• Edge-case videos (with privacy-safe handling)
• Simple improvements that increase reliability

Recommended: Apache 2.0 (permissive, safe for broad adoption).