video2images

Converts a video into images filtered for Photogrammetry.

Its main goal is to reduce the on-site time needed for data capture and also the time for photogrammetry software, such as Meshroom, to compute by only using a small selection of frames from a video data capture. This tool filters out frames that are too blurry to be used and consecutive frames that have too many features in common.

Table of Contents

• video2images
  • Table of Contents
  • Getting Started
    • Installation Prerequisites
    • Usage
  • How It Works
  • Acknowledgments

Getting Started

Installation Prerequisites

• Git
• VSCode
• Python
• Poetry

To install Poetry, follow these steps:

python -m pip install --upgrade pip
python -m pip install --user pipx
python -m pipx ensurepath
pipx install poetry

To install the project dependencies, execute the following command:

poetry install

To activate the virtual environment, run:

poetry shell

Usage

For command line arguments, use:

python video2images.py --help

To convert a video into images, run:

python video2images.py PATH_TO_VIDEO

This command will create a folder named extracted_frames at the same location as PATH_TO_VIDEO.

How It Works

The process is based on a simple, pipeline-oriented approach. No multiprocessing, threads, or clusters are used.

The pipeline operates as follows:

1. Create an output directory or clear if already existent
2. Open the video file and retrieve needed properties
3. Processes each frame with the following quality checks:

   1. Blurriness Check:

      Filters out frames that are too blurry based on the blur_min_threshold.

   2. Feature Check:

      Filters out frames that do have too many features in common with its previous frame based on feature_max_threshold.

4. Save Valid Frames
5. Cleanup & Print Stats

Acknowledgments

Special thanks to the following resources for their contributions:

• Adrian Rosebrock's Tutorial for providing valuable insights on detecting and debugging blur in frames.
• AzureDVBB for demonstrating an alternative approach to filtering frames from a video. For additional context, see the related article, Photogrammetry Datasets from Video - A Slightly Less Naive Approach.

If you have other possible approaches or suggestions, please contact me at robert.huem@gmail.com.