Tool build to generate and edit training dataset for sport activity recognition AI.

This project is a Python3 application designed to generate the training dataset used by AI for automated sport activity recognition.

It ensures the following features :

1. Synchronization of the raw data and video recorded during an experimentation.
2. Mark the event seen on the video to tag the raw data.
3. Export the tagged data as a standard training dataset.

This application has been built to run on multiple operating system and the installation procedure differ from one to another.

To install the application you will have to follow one of these guide :

• Windows Installation Guide
• Ubuntu Installation Guide

The application is quite intuitive and easy to use, as a result this part will be focused only on the main operations.

First of all you have to conduce some experimentation in order to obtain real data and the associated video feed.

Then, you will have to standardize the raw data to match the requirements to use them with the application.

Any video file format can be used with this application as long as the video codec has been installed on your computer. The following file format are the most likely to be installed by default on your operating system : .AVI, .WMV, .MOV, .MP4.

1. On Windows There are two way to start the application under the Windows operating system, but you are most likely to use the first one that is more convenient.

The first method consist of opening the application by double-clicking on the 'app.py' present in the installation folder. This should launch the application in matter of seconds, but if nothing shows up or if a text editor you should use the second method.

The second method of opening the application is less practical, but as functional as the first one. First of all, open the installation folder of the application in the file explorer. Then , press on the 'CTRL' key and right click on a blank spot in the folder view and click the 'Open a PowerShell here'. Finally, type the following line in the new (blue) opened window : python3.exe app.py

2. On Ubuntu Unfortunately, there is only one way to open the application using ubuntu.

Open your favorite terminal emulator and go in the installation folder of the application. Then, launch it by typing $: python3 app.py, or if you set the right file permission $: ./app.py.

The raw data must be standardized and aggregated in a single CSV file in order to be used with this application.

This file must contains at least one of the following parameter in order to synchronize the raw data and the video feed :

• ACC_X -> The accelerometer X axis.
• ACC_Y -> The accelerometer Y axis.
• ACC_Z -> The accelerometer Z axis.

You can also use a previously generated training dataset if all the previous requirements are fulfilled.

Example of input data :

Advanced example are also available in the demo_data directory.

• Lucas GRELAUD - LIARA Intern (2018) - Github

This project is licensed under the GNU GPL v3.3 - see the [LICENSE][gnu-license] file for details.