Dynamic-Hand-gestures-Recognition

Computer become a necessity and critical part of our daily life. Human and computer interaction is carried out through traditional input tools such as the mouse, keyboard and other conventional

• [ ] IMPORTANT NOTES

• The technologies used:

  Python 3.7 TensorFlow GPU V2 KERAS OpenCV pynput

• [ ] Description to read before you begin:

• extrecting_training_samples.PY: due to the limitation of the processing power this script is used to extract wanted gestures and the num of samples per gesture.
• test_app.py: is used to start the cam and predict gesture
• controll_commands.py: the script is used to link between the gesture predicted and the command to get applied
  • train.py: to train models
  • plots_traindata.py: is imported while taring to plot training resutls
  • training_samples folder : -containg subset form 20 bn jester data set which is used for the training purpose
  • resultfolder : after traing the code genrates txt file contains information about every epoch in the traing proccess
  • saved_modelfolder : usign callbacks ,best models wiehgts are saved into this folder while training

the subset of the 20 bn jester data Dataset link in oder to rain the model.

How to run the code to control Spotify :

• make sure Spotify app is already downloaded if not download it using windows 10 Microsoft store
• run test_app.py
• wait until the webcam launch
• open Spotify (note make sure Spotify is selected and in full-screen mode)
• make sure you click on the area of spotify controlers on the botthom

Experiment results

• The model is trained on training sample contains 1800 video clips for 5 different gestures.
• Model is a combination network between 3D CNN and LSTM
• The gestures are combination of 2 static gestures and 3 dynamic gesture.

• The model achieved 98% on the training accuracy and 90% on the validation accuracy.

  Accuracy

Loos