Endymion

An interactive video installation in which a viewer's stillness shapes a projected image — and the image, in turn, shapes the viewer.

First exhibited: 21 May 2026
Webpage: reedobeirne.com/endymion-revisited

---

What it is

A single viewer enters a small, cave-like space and sits in a chair facing a projected film. The system watches them with a Kinect depth sensor. When the viewer is still, the image brightens slowly (like moonlight appearing). When the viewer movse, the image retreats into darkness and holds there before gradually recovering.

The piece is named after the Greek myth of Endymion: the sleeper in a cave on Mount Latmos who sleeps with open eyes while Selene, goddess of the Moon, descends each night to watch him watch her in his stillness. The sensor sees the viewer in ways the viewer cannot return: depth, velocity, silhouette. This asymmetry is part of the work.

Stillness is the input. The system rewards sustained attention, not gesture.

More info: https://www.reedoco.io/endymion-revisited/

---

Hardware

Component	Role
Raspberry Pi 5 (8 GB)	Production runtime — runs the full stack at 23.976 fps
Kinect v1 (Xbox 360)	Depth sensor — background subtraction + foreground mask differencing
Nebula Mars 3 Air projector	1080p display, 400 ANSI lumens
MacBook	Development, SSH access, log analysis

---

Software stack

• Python 3.11 — main application
• libfreenect / freenect — Kinect v1 depth + IR data
• OpenCV — depth processing, foreground extraction
• pygame — video playback and display loop
• NumPy — frame-level image processing

---

How it works

[Kinect depth sensor]
        │
        ▼
[KinectSensor] — background subtraction → foreground mask → frame-to-frame pixel diff
        │
        ▼
[PenaltyRoutine] — three-phase presence signal: FALLING → HOLDING → RISING
        │
        ▼
[Effects pipeline] — luminosity · vignette · desaturation applied to video frame
        │
        ▼
[pygame display] — 1920×1080 fullscreen at 23.976 fps

The sensor is a velocity detector, not a position detector. It measures how many foreground pixels changed state between consecutive frames. Slow movements (breathing, micro-shifts) produce little frame-to-frame change and go undetected. Fast movements (fidgeting, arm gestures, entry and exit) produce large change counts and trigger the penalty. This is intentional: the piece rewards body-level commitment to stillness.

The presence signal follows a three-phase routine:

• FALLING — motion detected; image snaps toward darkness using a fast time constant
• HOLDING — viewer becomes still, but holds at nadir for a fixed duration (the penalty is unconditional)
• RISING — slow climb back to full brightness over several seconds

---

Running it

Development mode (no hardware required)

git clone https://github.com/Great-Bucket/endymion-public.git
cd endymion-public
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp env.example .env

Run with the mock sensor (simulated sine-wave presence signal):

SENSOR_TYPE=mock FULLSCREEN=0 python main.py

Run with a webcam (frame differencing, no Kinect needed):

SENSOR_TYPE=camera FULLSCREEN=0 python main.py

The repo includes a 1-minute sample clip of the installation film. Set in .env:

VIDEO_PATH=assets/video/1min-slowFilm_10mbps_endymion-installation_h264.mp4

On Raspberry Pi with Kinect

sudo apt install freenect libfreenect-dev
pip install freenect
SENSOR_TYPE=kinect FULLSCREEN=1 python main.py

See docs/operations/HANDOFF-pi-setup.md for full Pi setup, SSH configuration, and calibration procedure.

---

Key parameters (.env)

Variable	Default	Effect
SENSOR_TYPE	camera	mock · camera · kinect
VIDEO_PATH	(sample clip)	Path to the video loop
VIDEO_SPEED	1.0	Playback speed multiplier (0.5 = half speed)
EFFECTS	luminosity	Comma-separated: luminosity,vignette,desaturation
KINECT_MASK_CHANGE_THRESHOLD	40000	Motion sensitivity — calibrate on site
HOLD_DURATION	2.4	Seconds frozen at nadir after last motion
RECOVERY_DURATION	8.0	Seconds to climb back to full brightness
EMPTY_CAVE_DARK	1	1 = dark when empty (production); 0 = bright when empty (legacy)
DEBUG	0	1 = print sensor values to terminal

---

Documentation

The docs/ folder contains the full engineering record of the project:

Doc	What it covers
docs/reference/Project-Context.md	Conceptual framing, design principles, hardware inventory
docs/reference/DECISIONS.md	Key architectural decisions and the reasoning behind them
docs/history/CALIBRATION-log.md	Sensor algorithm iteration (v1 → v2 → v3), parameter search, key findings
docs/history/SOAK-TEST-REPORT-20260514.md	6-hour stability test, failure analysis, IR drift theory
docs/history/EXHIBITION-REPORT-20260521.md	First exhibition: what worked, what failed, lessons for next time
docs/operations/AT-VENUE.md	Step-by-step at-venue setup and calibration
docs/engineering/	Technical specs: presence signal, effects performance, monitoring, background drift recovery

---

License

Code: CC BY-NC 4.0 — use and adapt with attribution, non-commercial only.
Video and images: CC BY-NC-ND 4.0 — share with attribution, no modifications, non-commercial only.
© Reed O'Beirne. See LICENSE for full terms.