This project implement a neural network trained using tensorflow
library with python 3.8.10
. It is adapted to thermal sensor, with the aim to detect when a human being is in front of the captor.
This neural network should be able to run on a STM32F4 board.
Run the setup
script to install the dependencies for this project :
./setup
The project can be tested by running the demo script from the root directory :
./demo
It will get all the raw data from the data/demo_dataset.txt
input file and make a prediction wether there is a human shape
in them or not and print the result in the terminal. For validation purpose, all the specified thermal frames will be displayed as pictures (heatmap) into the data/demo_heatmaps
dir.
You can train a model to detect human shapes from thermal sensor data if the data is formatted as the files presents in the data
directory.
Just cd
into the src
directory and run the train_model
using python > 3.8
.
The model is fully customizable by changing the lines in the config.py
file !
Suit yourself.
Also, you can automatically tune the hyper-parameters by running the tune_parameters.py
file.