Our eyes are continuously fixed on the computer monitor during the day and occasionally shift to the smartphone, followed by eye grabbing shows on television. This constant strain and stress has given rise to eye problems. Eye Exercises app is a good way to periodically maintain eyesight and vision. It is useful for reducing the symptoms of numerous eye disorders (eye problems) like myopia (nearsightedness), hyperopia (farsightedness), astigmatism or eye straining.

The goal of the Eye Exercise Application is to help increase effectiveness of eye exercise. We detect eye gaze of a user using computer vision and pretrained machine learning models. We check it with eye gaze coordianates of instructor and show user a live score of exercise effectiveness. We also want the exercise assistant to be voice enabled to instruct and motivate users for exercise.

We have used four pretrained machine learning models from OpenVINO™ toolkit, short for Open Visual Inference and Neural network Optimization toolkit:

1. Face Detection: Detects face coordinates from video or webcam images
2. Head Pose Estimation: Detects pose coordinates for head
3. Facial Landmarks Detection: Gives coordinates or location for facial landmarks like eyes, nose and mouth
4. Gaze Detection Model: Takes head pose coordinates and eye landmark as input and predicts gaze vector

We compare eye gaze vector of instructor and User using cosine similarity.

>> from scipy.spatial.distance import cosine
>> eye_gaze_instructor = [ 0.62916321,  0.10232677, -0.77875257]
>> eye_gaze_user = [ 0.09647849,  0.03398839, -0.82852501]
>> cosine(eye_gaze_instructor, eye_gaze_user)
0.15561332345537238

Step 1. Download below three tools:

1. Microsoft Visual Studio with C++ 2019, 2017, or 2015 with MSBuild *If you want to use Microsoft Visual Studio 2019, you are required to install CMake 3.14.
2. CMake 3.4 or higher 64-bit
3. Python 3.6.5 64-bit

Step 2. Download OpenVINO™ toolkit 2020.1 with all the prerequisites by following this installation guide

Step 3. Setup OpenVINO™ toolkit 2020.1 using the commands below in command prompt

cd C:\Program Files (x86)\IntelSWTools\openvino\bin\
setupvars.bat

Step 4. Configure model optimizer using the commands below in command prompt

cd C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\model_optimizer\install_prerequisites
install_prerequisites.bat

Step 5. Verify installation

cd C:\Program Files (x86)\IntelSWTools\openvino\deployment_tools\demo\
demo_squeezenet_download_convert_run.bat

Above commands should give output like this image

Step 1. Clone the repository using git clone https://github.com/bhadreshpsavani/EyeExerciseApp.git

Step 2. Create virtual environment using command python -m venv base in the command prompt, then activate environment using the commands below,

cd base/Scripts/
activate

Step 3. Install all the dependency using pip install requirements.txt.

Step 4. Instantiate OpenVINO™ toolkit environment. For Microsoft Windows use the commands below

cd C:\Program Files (x86)\IntelSWTools\openvino\bin\
setupvars.bat

Step 5. Go back to the project directory src folder

cd path_of_project_directory
cd src

Step 6. Run the commands below to execute the project

python main.py -fd ../intel/face-detection-adas-binary-0001/FP32-INT1/face-detection-adas-binary-0001.xml -lr ../intel/landmarks-regression-retail-0009/FP32-INT8/landmarks-regression-retail-0009.xml -hp ../intel/head-pose-estimation-adas-0001/FP32-INT8/head-pose-estimation-adas-0001.xml -ge ../intel/gaze-estimation-adas-0002/FP32-INT8/gaze-estimation-adas-0002.xml -i cam

Command line argument information:

• fd : Specify path of xml file of face detection model
• lr : Specify path of xml file of landmark regression model
• hp : Specify path of xml file of heead pose estimation model
• ge : Specify path of xml file of gaze estimation model
• i : cam for Webcam

bin: This folder has demo.mp4 file which we are using for eye excercise video

imgs: It contains images used in this project for documentations and results

intel: This folder contains machine learning models in IR(Intermediate Representation) format

src: This folder contains model files, pipeline file(main.py) and utilities

• model.py is the model class file which has common property of all the other model files. It is inherited by all the other model files This folder has 4 model class files that has methods to load model and perform inference.
  • face_detection_model.py
  • gaze_estimation_model.py
  • landmark_detection_model.py
  • head_pose_estimation_model.py
• main.py file used to run complete pipeline of project. It has object of all the other class files in the folder
• input_feeder.py is utility to load local video or webcam feed

• Computer Vision
• Openvino Toolkit
• Python
• Microsoft Azure for Deployment(Future Step)

• Reduce lag between Instructor and User Video
• Handle Different different lighting conditions
• Increase Inference Speed

• Bhadresh Savani
• Pakeeza
• Erin Song
• Richa
• Jose Mariscal

• Create end to end pipeline to extract eye gaze coordinates
• Create pipeline for getting eye gaze coordinates for excersice video
• Create pipeline and UI for webcam video
• Develop score computation logic
• Develop UI for computer to view output and score
• Enable app with voice assistance
• Deploy application to Microsoft Azure Cloud
• Create a mobile app

• This repository is taking codes for detection and Gaze Prediction from my earlier project Computer Pointer Controller.
• Few images and Video file is taken from Udacity Intel Edge AI Nenodegree Program Project 3 (Computer Pointer Controller)