evio-evlib – Event Camera Detection Workbench

Nix-powered toolkit for fan RPM and drone detection using event cameras.

Fast windowing with evlib + Polars, fullscreen MVP UI with pink Y2K branding, reproducible environment.

Crafted for the Junction 2025 Sensofusion challenge by team weUseNixBtw — Matti Parkkila (Team Lead), Irene Bandera Moreno, and Jesse Karjalainen.

Highlights

• A unified GUI for exploring event-based camera data
• A preprocessing pipeline to transform proprietary binary data from .dat to industry-standard .h5
• A blazingly-fast event-data loader (yes, python bindings with rust backend)
• A robust, multi-layered computer vision algorithm for real-time RPM counting & drone detection
• All packaged in an elegant flake.nix - just "nix develop" and "run-mvp-demo"

Quick Start

nix develop
run-mvp-demo  # Fullscreen UI with fan/drone detectors

What Works

• 🟢 Detectors: Fan RPM (ellipse + DBSCAN), Drone propellers (multi-ellipse)
• 🟢 MVP UI: Fullscreen launcher, 0.25x-100x playback, 10μs-100ms windows
• 🟢 Performance: Lazy loading, HDF5 chunking, frame skipping for 6000+ FPS
• 🟢 Visual: 870x435 thumbnails, Sensofusion gray + Y2K pink palette

Commands

run-mvp-demo               # Menu UI with all detectors/datasets
run-fan-rpm-demo <file>    # CLI fan RPM (evlib/Polars)
run-drone-detector-demo <file>
convert-all-legacy-to-hdf5 # Convert .dat → .h5 for evlib

Roadmap

• 🔴 Stream adapter (Metavision SDK) for live cameras
• 🔴 Storage sink (ClickHouse/TSDB) for telemetry
• 🟡 RVT integration (runs, retraining incomplete)

Docs (docs/)

1. Problem statement
2. Solution approach
3. Architecture (mermaid diagram)
4. Tech stack & commands
5. Repo structure
6. Status (green/yellow/red)
7. Runbook (data prep, demos)
8. Data conversion
9. Detector pipelines
10. RVT proof of concept