Edge Impulse TMF8829 training and live classification interface

Implementation of a Edge Impulse interface for

• Training of AI models
• Running live classification

This interface operates on TMF8829_EVM_DB_DEMO or TMF8829_EVM_EB_SHIELD boards.

Example classification for cup detection, proof of concept, where TMF8829 is located 30 cm above the desk looking downwards - the AI model is available at Edge Impulse website: TMF8829_48x32_cup_detection:

Setup

• TMF8829 EVM connected to PC - TMF8829_EVM_DB_DEMO or TMF8829_EVM_EB_SHIELD
• Edge Impulse account

Create an API Key in https://studio.edgeimpulse.com/

Goto Project Dashboard -> Keys -> New API keys, create a key and copy the key number from the browser to tmf8829_edge_impulse.py

API_KEY = "ei_<insert_key_here>"

For live classification to work, add a HMAC Keys as well. This key needs to have 'Set as development key' checked. The API_KEY above is not affected by the HMAC key and there is no need to copy it to these sources.

Installation

Virtual environment

Recommendation is to set-up a virtual environment. Open your favourite Windows PowerShell, VisualStudio Code etc. To install a virtual environment named env, and use it:

python -m venv env
# If PowerShell blocks the script, run: Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope Process
./env/Scripts/Activate.ps1

Install libraries

Python version 3.10.11 or higher is required.

To run the scripts in this folder you need to install the packages in the requirements.txt file with:

pip install -r requirements.txt

All required python packages are inside the subdirectory packages.

Usage

If you are using TMF8829_EVM_EB_SHIELD, start tmf8829_zeromq_server.py first; this can be done with the pre-compiled server file from TMF8829_Driver_ZMQ_Server_Client_EXE_<latest version>.zip or inside a separate shell

python tmf8829/zeromq/tmf8829_zeromq_server.py

If you are using TMF8829_EVM_DB_DEMO, no additional server needs to be started.

Training data

To collect training data, update label as needed in tmf8829_edge_impulse.py

# Set Training Label
TRAINING_LABEL = 'Empty_cup'

To start collecting data, run tmf8829_zeromq_training_client.py

python tmf8829/zeromq/tmf8829_zeromq_training_client.py

This will automatically copy the collected and labeled training data to the Edge Impulse project defined by the API_KEY.

Live classification

First start the client tmf8829_live_classification.py

python tmf8829/zeromq/tmf8829_live_classification.py

Go to the Edge Impulse project and connect to the TMF8829 device:

Visualization on the PC

The EVM GUI can be used in parallel to this application, but needs to be started AFTERWARDS.

Configuration

Update file cfg_client.json with following examples:

• Parameter period [in ms] to modify speed of detection.
• Parameter iterations [in k iterations] is used to change performance of detection

Info

This is a fork of tmf8829_driver_python modifying files to create an application, which can run together with TMF8829_EVM_DB_DEMO or TMF8829_EVM_EB_SHIELD.