ImageFusion AI — RGB & Thermal Calibration Hub
An advanced industrial web platform designed to align, fuse, and calibrate multimodal image pairs (RGB Visible Light and Thermal Infrared) with GPU-accelerated previews and optimized grid search.
About the Project
ImageFusion AI is an advanced, high-performance web dashboard that replaces heavy desktop GUIs with a premium light-glassmorphic Single Page Application (SPA). It integrates computer vision edge alignment, automated coordinate grid search, and high-concurrency background batch processing queues to align and fuse infrared thermographic views with color visible-light images.
Core Calibration & Fusion Pipeline
To bridge the modal gap between heat signatures and color photos, the alignment algorithm analyzes structural geometries rather than pixel color matching:
Raw Inputs
RGB & Thermal Pairs
Sobel Edge
Gradient Filter
ROI Masking
Noise Filtering
Grid Search
Coarse-to-Fine Opt
Fused Output
Affine Warped Blend
Key Features & Optimization Challenges
Designed for sub-second, hardware-efficient rendering under cloud container restrictions:
Split-Screen Swipe View: Swipe-to-reveal divider layout that allows inspectors to wipe back and forth across structural boundaries.
Coarse-to-Fine Grid Search: Searches translation and scaling parameters recursively to maximize boundary overlap alignment scores.
Workspace Downscaling (60x Speedup): Scales internal optimization frames from 1024x768 to 256x192, slashing processing time from 60 seconds to under 0.94 seconds without accuracy loss.
Client-Side CSS Transforms (0ms Latency): Updates slider changes instantly in the viewport via GPU-accelerated CSS transforms and opacities, offloading heavy OpenCV decompressions from the backend.
Batch Automation: Serves concurrent background tasks via FastAPI to scan directories and output fully fused frames automatically.
Tech Stack & Frameworks
Built upon high-speed computer vision modules and async web architectures:
Backend APIs: FastAPI (high-performance async web framework), Uvicorn ASGI server
Image Processing Library: OpenCV (cv2) for affine transformations and edge analysis
Array Optimization: NumPy (matrix calculations and region indexing)
Validation Layer: Pydantic schemas for data safety
Frontend: HTML5, Vanilla CSS3 (Light Glassmorphism variables, absolute stacking grids), Vanilla JS (Fetch API, drag physics, API rate throttling)
Project Demo
Project Note
🚀 Industrial Drone Inspection Applications This project aligns Parallax Shifts between physical drone lenses (RGB and Long-Wave Thermal Infrared sensors), enabling operators to accurately map temperature anomalies (e.g. electrical hotspots, fluid pipeline leaks) onto visual components.