Athlete-Spy

Human pose estimation tool using TensorFlow Lite and OpenCV

A simple commandline tool for extracting and saving keypoints of a human pose from videos.

(C) Copyright Nico Mahler 2024

Prerequisites

• Python v 3.4+

    $ python --version
  

• CMake v 3.15+

    $ cmake --version
  

• Conan package manager

    $ pip install conan
  

  (Hint: If not installed globally, add the scripts directory to path enviroment variable)

    %USERPROFILE%\AppData\Local\Packages\%PYTHON_VERSION%\local-packages\%PYTHON_VERSION%\scripts
  

Installation

• Clone repository

    $ git clone https://github.com/Phlegmati/Athlete-Spy
  

• Create conan build profile

    $ conan profile detect
  

• Installing dependencies with Conan

    $ conan install . -s compiler.cppstd=17 --build=missing
  

• Building

    $ cmake --preset conan-default
    $ cmake --build --preset conan-release
  

Usage

• Run build with examples

    $ build\Release\athlete-spy.exe
  

  (Hint: Looks for example files in the "asset" folder in the current runtime enviroment)

    athlete-spy
    ├── assets
    │   ├── lite-model_movenet_singlepose_lightning_tflite_float16_4.tflite
    │   └── [email protected]
    └── build
        └── Release
            └── athlete-spy.exe
  

• Manual Page

    $ build\Release\athlete-spy.exe --help
  

• Pass video file (supported: *.mp4 *.avi)

    $ build\Release\athlete-spy.exe --video=path/to/video.file
  

• Pass model file (supported: *.tflite)

    $ build\Release\athlete-spy.exe --model=path/to/model.file
  

Output

• The processed pose video will be saved as an *.avi video and *.txt file

    athlete-spy
    └── output
        ├── video_processed.txt
        └── video_processed.mp4
  

References

• See documentation of the used Model https://tfhub.dev/google/lite-model/movenet/singlepose/lightning/tflite/float16/4

• See documentation of the used TFLite API (v. 2.12.0) https://www.tensorflow.org/lite/api_docs/cc

• See documentation of the used OpenCV API (v. 4.5.5) https://docs.opencv.org/4.5.5/