Signed

Real Time Translation of Sign Language to English.

Setting Up the environment.

• Set up an anaconda environment as Python 3.5.6, and activate this environment.
• Use command prompt or anaconda prompt to install libraries by using install_packages.txt using pip. Note: Use command prompt or anaconda prompt to run dependencies.py. If this runs without any trouble we can now start running our code. This step is necessary because on a lot of systems tensorflow is not installed properly with pip installation on the first go. For my particular environment I used pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-2.0.0-cp35-cp35m-win_amd64.whl as specified on the Tensorflow website.

Creating your own custom dataset.

• Run train.py to create your own dataset.
• When the Histogram screen pops up, cover the small green boxes with your palm completely and press c. Your palm should be white and the background should be black.
• Press s to save the histogram and stop recording.
• Input the Hand Sign Label you want the system to store, and input the no of training images.
• If the label of your dataset is Ex, the images are stored under .\signs\train\Ex\. Some example thresholded images are shown below.

Predict hand-signs in real time.

• Run predict_hand_sign.py to start predicting handsigns.
• When the Histogram screen pops up, cover the small green boxes with your palm completely and press c. Your palm should be white (in the foreground) and the background should be black.
• Press s to save the histogram and stop recording.
• Put your palm inside the green box and start 'signing'. The prediction will be shown on the screen.