ProxErgo

Posture & Fatigue Monitoring

A Privacy-First Posture & Fatigue Monitoring System
Ergonomic assessment on any laptop with a front-facing camera

---

Overview

ProxErgo monitors posture and fatigue using your front-facing webcam. Built on clinically-validated biomechanics and privacy-by-design: only landmark coordinates are processed, never raw video.

Key Features

Traditional Approach	ProxErgo
Requires side-view camera setup	Works with any front-facing webcam
Expensive clinical equipment	Runs on any laptop
Data sent to cloud servers	100% local processing
Generic "slouch detection"	CVA proxy + depth + vertical metrics
Single metric	Multi-factor posture + fatigue + sway

---

Project Structure

ProxErgo/
├── Clinical_Research/          # Evidence-based thresholds
│   ├── clinical_thresholds.py
│
├── Model_Development/          # Core analysis
│   ├── core/
│   │   ├── biomechanics.py      # CVA, depth, vertical, tilt, sway
│   │   ├── fatigue_detector.py # PERCLOS & blink detection
│   │   └── posture_classifier.py  # XGBoost + rule-based fallback
│   ├── detectors/              # MediaPipe pose & face mesh
│   └── training/               # Data collection & model training
│
├── Streamlit_App/              # Web dashboard
│   ├── app.py
│   └── components/
│
├── config.yaml                 # Thresholds & settings
├── demo_opencv.py              # Standalone OpenCV demo
├── environment.yml
└── requirements.txt

---

Quick Start

1. Set Up Environment

conda env create -f environment.yml
conda activate proxergo
# or: pip install -r requirements.txt

2. Run the Dashboard

streamlit run Streamlit_App/app.py

3. Or Run the Standalone Demo

python demo_opencv.py

Demo controls: q Quit · c Calibrate · s Skeleton · m Metrics · r Reset

---

Calibration

Calibration personalizes the baseline for your posture and setup:

1. Press c (demo) or click Calibrate Posture (app)
2. Sit in your best upright posture
3. Hold still for ~30 frames (~1 second)

Calibration captures your neutral depth, vertical ear–shoulder distance, and shoulder tilt. When calibrated, the system uses your baseline instead of generic thresholds. Recalibrate when you change clothing or camera position.

---

Posture Classification

7-class system: optimal, forward_head_mild, forward_head_severe, tilted_left, tilted_right, too_close, too_far.

A forward-head score combines depth, vertical ear–shoulder distance, CVA, and slouch. When tilted, vertical/CVA are excluded (they invert).

Score	Classification
< 0.25	Optimal
0.25 – 0.65	Mild forward
≥ 0.65	Severe forward

Tilt overrides forward head when shoulder tilt deviation exceeds 7° (calibrated) or 10° (uncalibrated).

---

Key Metrics

Posture

• CVA proxy: arctan2(|ear_x − shoulder_x|, shoulder_y − ear_y)
• Depth: nose vs shoulder midpoint (positive = forward)
• Vertical: shoulder_mid_y − ear_mid_y (head drop)
• Slouch: max(0, expected_height − nose_y) / shoulder_distance

Fatigue (PERCLOS)

Level	PERCLOS	Action
Alert	< 15%	Normal
Mild	15–25%	Monitor
Moderate	25–40%	Take a break
Severe	> 40%	Rest needed

Postural Sway

Tracks nose position over time. Amplitude, path length, velocity, and frequency indicate neuromuscular fatigue.

---

Privacy by Design

• Raw video frames are never saved
• Only skeleton coordinates are processed
• All inference runs locally
• No network transmission of posture/fatigue data
• Calibration stored in data/calibration.json (local)

---

Report

See Clinical_Research/Fatigue Tracker ProxErgo Report.pdf for the report.

---

ProxErgo · An expert exists locally on your device.