Graduation Project | CS-12

I Hear U, precedes a way for sign language recognition using computer vision technology to translate signs into Arabic language.
Demo video: https://youtu.be/KH9pcdHTJOA
Project documentation: http://surl.li/zicz

How to setup:

• clone the repo git clone https://github.com/Checkmate-2/IHearU.git
• move to the the folder cd .\IHearU\
• creat virtual environment python -m venv venv
• activate the virtual environment venv\Scripts\activate.ps1
• install the requierments pip install -r requirements.txt

How to run:

• open the terminal in the code folder
• activate the virtual environment venv\Scripts\activate.ps1 if usuing powersell or venv\Scripts\activate.bat if using cmd
• run the code u want python collect.py python train.py python test.py

Important

• all data and model folders should be in the same folder as the code

1- Collect parameters

• data folder : where u wanna save the action
• action name : name of the action (a ,b , hello)
• number of sequences : number of data samples u want to collect
• number of frames per sequence : every sample has ( 2 frames ,10 frames ,30 frames) of data
• you cam source number : you camera number usualy (0) or (1) or (2)
• recording starts after 10 seconds, after the recording is done u action will be saved in the data folder

2- Train parameters

• data folder name : the folder that has ur actions data
• number of epochs : how many epochs u wanna train the model
• model name : name of the model if u want to save it .
• u get a tflite model, a complete model, a csv file for the actions

3- Test parameters

• modelname = the model folder name
• number of frames the model recieve : how many frames ur trained model recieve
• accuracy threshold : take a value from 0 : 1
• get highest prediction in last ... : number of predictions to consider

Team Members: Ahmed Habeeb, Ahmed Fayed, Amany Sherif, El-said Gamal, Emad Mohamed, Eman Mohammed, Habiba Mohamed, Mohamed Zaki, Omar Alkholy, Shahenda Wafa

Under the supervision of: Dr. Mohammed Alrahmawy, Dr. Zahraa Tarek, Dr. Mohamed Handosa. Faculty of Computer & Information Sciences - Mansoura University