wonnor-pro / gesture_demo

Guidance of using the hand gesture demo

Dendencies and Environment

• Python 3.7
• OpenCV 4.1.0
• tensorflow 1.14.0

Model Path

the inference code has implemented two cascaded ML models. First, the MTCNN network and then a simple CNN network.

The network structure of MTCNN is defined in the Train_Model folder, while the CNN network structure is coded in the inference file.

MTCNN model weights are stored in the MTCNN_Model folder, P-Net, R-Net, O-Net separately. CNN model weights are stored in the CNN_Model folder, please refer to the checkpoint file in it.

If you want to change the path, just change the value of

model_path = ['MTCNN_Model/PNet/PNet-30', 'MTCNN_Model/RNet/RNet-22', 'MTCNN_Model/ONet/ONet-22']
model = load_classification('CNN_Model/kcnn')

inside the inference code.

Run camera inference

just run camera.py for hand detection (MTCNN only) or camera_c.py for hand gesture classification (both MTCNN and CNN).

hand gesture classes are defined below:

class_names = ['0SingleOne', '1SingleTwo', '2SingleFour', '3SingleSix',
               '4SingleEight', '5SingleNine', '6SingleBad', '7SingleGood']

The classification is not of good performance in this beta version.