Biomechanical Ergonomic Evaluator

A real-time posture analysis system that connects to an edge camera stream to evaluate your spinal alignment. It utilizes a 3D Human Mesh Recovery engine to track internal S-Curve angles and dynamically alerts you when your posture deteriorates.

Key Features

• 3D Mesh Tracking: Utilizes the ROMP (Robust One-stage Method for Person) library to calculate 3D joint rotations from standard video frames.
• Biomechanical Analysis: Calculates precise 3D angles to track Lumbar Slump, Thoracic Hunch, and "Text Neck".
• Dynamic Calibration: Asks the user to sit in a healthy pose upon startup, taking 20 samples to establish a personalized baseline average.
• Rolling Posture History: Maintains a sliding window of the last 15 frames to evaluate posture consistency and eliminate false positives.
• Asynchronous Audio Alarms: Triggers an alarm.wav background sound via Windows winsound if 8 out of the last 15 frames show a significant deviation from your calibrated baseline.
• Heads-Up UI Display: Overlays real-time system FPS, neck/mid-spine/lower-spine angles in degrees, and color-coded status alerts ("HEALTHY S-CURVE" vs "FIX YOUR POSTURE!") directly onto the video feed using OpenCV.

Project Structure

• main.py: The application entry point that handles the video capture loop, coordinates the evaluation steps, tracks the rolling 15-frame window, and triggers the audio alerts.
• core/config.py: Contains system constants such as the STREAM_URL and base geometric thresholds for the SMPL model.
• core/calibration.py: Collects initial pose samples to build a custom baseline array, enabling the system to track relative deviations (deltas) instead of static numbers.
• core/math_utils.py: Uses vector mathematics on specific SMPL mesh joints (pelvis, lumbar, thorax, upper_thorax, and neck) to compute the angle of curvature in a 3D space.
• vision/pose_engine.py: Encapsulates the ROMP neural network, which is configured specifically for real-time inference by turning off built-in rendering and focusing only on the primary person.
• vision/ui_renderer.py: Houses the OpenCV drawing functions to render the semi-transparent background panel and live metrics.

System Requirements

• Operating System: Windows (The system relies on the winsound module for asynchronous alarm playback).
• Hardware: An NVIDIA GPU is heavily recommended (Tested on RTX 3050) for real-time ROMP mesh extraction.
• Dependencies: opencv-python, romp, numpy.