Return to site

Complete Guide of CsvTool for Curie Action Learning/Recognition

A New Data Waveform Processing Application

· CsvTools,tutorial

Introduction

CsvTools is a graphical processing tool for Arduino/Genuino101 to conduct neural network learning/recognition, provide data sampling and motion data recognition.

Before that, we need to run csv file waveform processing in Excel, and each capture of the waveform data need to be saved as a separate csv file, then we put some files to the pipeline directory, configure the config file and run the pipeline.exe, and finally we were able to get the trained neuron.csv.

However, with CsvTools, we can streamline this process.

A. Preparation

Download the software:

After downloading, extract the zip file, then enter the directory and run the run.bat file to start the software.

When the software tool starts, you will see the following console information:

According to the prompt, enter http://127.0.0.1:3000 in the browser to enter the user interface:

B. Features

Below is a breif introduction of features of CsvTools' operation interface:

| Select file: select a ".csv" file in which the data is collected by Curie.

| Load: Re-import the data waveform.

| Delete the selection: use the mouse to select a region and delete the selected waveform data.

| Enter the zoom mode: use the mouse to select a region. There are two operations of the waveform. The default is to select the waveform; In the magnification mode, once again use the mouse frame to select a region, and the waveform will be enlarged. you can click the button " Select mode ".

| Save as a file: Save the waveform data as a file.

| Training recognition: After selecting, deleting, processing and saving of the data, the training recoginition will begin. Under the directory csvtools\pipeline-0.5, a neuron.csv file will be generated.

Boxing action recognition

A. Goal

1. We want to collect three kinds of boxing action through Curie: straight, swing, uppercut.

2. With CsvTools, we want to extract and analyse above three kinds of boxing data, and complete training.

3. We want to send the CsvTools generated data to Curie for action recognition.

B. Application

Select the DATA.csv which contains the collected data

After selecting, the data waveform will be automatically displayed:

Looking at the waveform, you can distinguish between three kinds of boxing action data area, as shown in the following Figure:

You can then use the "delete selection" button to remove unwanted waveform

The retained waveform data is as shown below:

Then click "Save File". Type file name "C-straight". The file contains data for the straight action.

Extract Swing data:

Extract Uppercut data:

After data extraction, you can go to the directory csvtools/pipeline-0.5/train_data to check the generated files:

Training recognition: Click on the "Training Recognition" button to carry out the action training (You'll be notified after completion). Then go to the directory csvtools \ pipeline-0.5, you will see that neuron.csv and the other two files, trainer_random_frames.csv ad trainer_source_frames.csv are generated; Finally, we will be able to install neuron.csv into Curie to identify the actions.

All Posts
×

Almost done…

We just sent you an email. Please click the link in the email to confirm your subscription!

OKSubscriptions powered by Strikingly