tower-vib

Active development. No license granted yet. Industrial-grade video vibrometry (WIP): ROI caching, phase-based subpixel displacement, and physics-oriented vibration features.

Overview

This repo measures transmission tower vibration from video.
Current pipeline:

1. Extract and cache a golden ROI video clip (default 200 frames) to .npz
2. Run a lightweight ROI mean-curve sanity check and export .png/.csv

Repo structure

• data/ : input videos (e.g. demo_tower.mp4)
• scripts/
  • extract_roi.py : cache ROI frames to results/cache/*.npz
  • roi_mean_curve_check.py : compute ROI mean curve and save plot/csv
  • run_pipeline.py : end-to-end runner (extract ROI + mean curve)
• results/
  • cache/ : cached ROI .npz
  • plots/ : mean curve plot .png
  • data/ : mean curve .csv
  • logs/ : run logs

Quick start

1) Install

pip install -r requirements.txt

2）Run pipeline(cache ROI + mean curve)

python scripts/run_pipeline.py

Excepted outputs

After running,you should get:

• results/cache/YYYY-MM-DD_main_roi_frames.npz

• results/plots/YYYY-MM-DD_mean_curve.png

• results/data/YYYY-MM-DD_mean_curve.csv

• results/logs/*.log

NPZ format

results/cache/*_main_roi_frames.npzcontains:

• frames:ROI frame sequence

  • shape:(T,H,W,3)for color ROI (OpenCV default BGR)

  • (optional future)(T,H,W)for grayscale ROI

• roi:ROI coordinates(x,y,w,h)

• video:source video path

• fps:frames per second

Notes

• ROI mean curve is used as a sanity check:

  • verifies ROI extraction correctness

  • help detect abnormal frames via sudden instensity jumps

• ROI is currently set inscripts/run_pipeline.pyas(x,y,w,h)