Visual Coherence Oracle

A lightweight motion-induced sparse representation system for epistemic validation of multi-sensor environmental anomalies on Raspberry Pi 5B.

Overview

This system generates a sparse, motion-based representation to validate whether multi-sensor anomalies (IMU, acoustic, atmospheric) correspond to real environment-level structural responses. The camera is NOT a detector—it's a validator.

Quick Start

# Install dependencies
pip install -r requirements.txt

# Run the visualizer demo
python demo/visualizer.py

# Use in your code
from src.agent.coherence_oracle import CoherenceOracle

oracle = CoherenceOracle(camera_id=0)
signal = oracle.get_coherence_signal()
print(f"Scene coherence: {signal.scene_coherence:.2f}")
print(f"Disturbance type: {signal.disturbance_type}")

Architecture

Camera Frame (720p)
    ↓
FAST Feature Detection (500 keypoints)
    ↓
Lucas-Kanade Optical Flow
    ↓
Motion Vector Field (8x6 grid)
    ↓
Spatial Coherence Analysis
    ↓
CoherenceSignal → Agentic AI Layer

Key Outputs

Metric	Range	Meaning
scene_coherence	0.0-1.0	Correlation of motion across spatial cells
motion_intensity	0.0-1.0	Normalized motion magnitude
disturbance_type	NONE/LOCALIZED/SCENE_LEVEL	Classification of motion pattern
visual_confidence	0.0-1.0	Trust in the visual assessment
false_positive_risk	0.0-1.0	Likelihood other sensors have false positive

Performance

• Target: <20ms per frame
• Processing rate: 15 FPS
• Tested on: Raspberry Pi 5B

Project Structure

Camera/
├── src/
│   ├── config.py           # Configuration constants
│   ├── camera/             # Frame capture
│   ├── motion/             # Feature detection & optical flow
│   ├── representation/     # Sparse coherence grid
│   └── agent/              # Oracle interface
├── tests/                  # Unit tests
├── demo/                   # Visualization tools
└── requirements.txt