This is a working area for all the software for Hephaistos.

Sign Language dataset: https://imperiallondon-my.sharepoint.com/:f:/r/personal/at2123_ic_ac_uk/Documents/Hephaistos?csf=1&web=1&e=c9XmRr

#---------------------------------------------#

INSTRUCTIONS

#---------------------------------------------# To reproduce:

1. Run the file 'data_recorder.py' in the 'mediapipe' folder. This will access the camera on your device and overlay the MediaPipe landmarks live.

2. Stay within the live feed and press 'r'. You will see a countdown on your screen. A recording, and its corresponding file of landmarks, will be made. The video will be saved in the 'recorded_videos' folder. The landmark data will be saved within the 'landmark_data' folder.

3. Press 'q' to quit the screen.

4. You can alter where your files are saved by changing the following variables in the 'DIRECTORY AND FILE SET-UP' in 'data_recorder.py': video_dir = "recorded_videos/(label name)" landmark_dir = "landmark_data/(label name)" video_title = "video_(label name)(recording number).mp4" landmark_title = "landmark(label name)_(recording number).npy"

5. Once all videos are recorded, go to the 'training' folder and run 'train_LSTM.py'. This will take care of the data and train the LSTM. A 'label_map.json' should be written.

6. After exporting the model from Google Collab, import the model back.

7. Go to 'detection.py' in the main directory. Update the variable 'MODEL_PATH' to where your model is.

8. Running 'detection.py' gives you real-time detection.

#---------------------------------------------#

THINGS TO BE AWARE OF

#---------------------------------------------#

1. The variables 'self.sample_interval' in hand_tracker.py and 'fps' in the main script of data_recorder.py SHOULD BE CONGRUENT.