Our main goal was to create a system for handsfree navigation. There were many ways that we considered, but the detection of eyes blinking and converting those to morse dot and dash and using different combinations of those to create different simulated keypresses seemed most intuitive.
We learned how to work with frames from videos and analyze them individually. Using python libraries like imutils and opencv we converted each frame into a gray image and calculated the change in pixel value on each edge and labeled each pixel value change and were able to recognize different parts of the face. Then taking each frame, we compared it with the trained model, and the output from that model was put into an array, and after waiting a bit, the array was compared with the morse codes and converted into the corresponding value.
It was hard to find the proper dataset with images of the eyes; our computational power was also too limited to take the dataset and train the model. Luckily, we found a trained model which could detect the face, and even after the trained model, our program needed optimal condition to run smoothly. We tried to make the program flexible by taking the input from a phone camera,but the lag between the data input from the external source and the data processing made it a little impractical for our current setting. Thus we got stuck with the laptop camera and we had to make it work.
We learned machine learning techniques to detect eye vision, learned how to decipher the morse code, and how to work together in a team.
If we can get more computational power, we can carry out our project even further and make it available in more systems, and possibly even through clouds in mobile devices.
Log in or sign up for Devpost to join the conversation.