Who Are We
Neuroworks is a company established in 2022 to help people with walking disabilities with the EEG system.
Our Mission
Helping people with walking disabilities.
Our Vision
HARDWARE DESIGN
At the beginning of this project, we were aimed to use fNIRS to read user’s brainwaves. After designing fNIRS based hardware, we noticed that fNIRS is not appropriate for our project. A hardware, that is EEG based, was seemed more approriate for our project. Then, we started to desing new hardware.
Visualization of EEG Data
After designing EEG circuit, we wrote an esp32 code to read data from EEG. But the data read from EEG was not meaningful. Then we used arduino board to read meaningful data.
Testing Hardware
Firstly, we plug arduino and EEG to each other. Then by connecting reference electrode to user’s ear and connecting the other two electrodes to user’s forehead, we got the data.
We ask user to
- think something and focus on,
- solve a mathematical problem,
- hold breath a couple of seconds than give,
- think about turning right and focus on,
- think about turning left and focus on,
- think about going forward and focus on,
- think about stopping and focus on,
then observed the difference on EEG data. Because of the noise, the data read from EEG was too sensitive. We could not solve this problem.
We used plot tool of arduino to visualize EEG data.
DATA ANALYSIS
We used an eeg dataset with 768 samples for data analysis. We analyzed the data under 4 different headings and classified these data by using the Random Forest Algorithm. During the classification process, 537 samples were used for training and 231 samples were used for testing. As a result of these operations, the algorithm gave the following outputs:
Forward Value : 4.69921875
Back Value : 163.015625
Left Value : 21.4921875
Right Value : 70.9375
Accuracy : 0.6103896103896104
SIMULATION
After training the model, the values that the we take (forward,back,right,left) are used in simulation. We assumed that the harware sends 4 different values corresponding to the right,left,forward and back. We calculate which direction is dominant by using values that we take from the trained model. After that, the true direction movement will performed. While performing the movement it will not take another command.
