Hand Gesture Detection Using VGG-16

This work, focusing on the VGG16 architecture, developed a CNN model for hand motion detection. Methods of data augmentation to enhance resilience were incorporated by a library of photographs of hand gestures as training for the model. The batch size used in training the Adam optimizer was thirty-two. After ten ten-epoch training, the model for validation arrived with about ninety-six percent accuracy. Two dropout layers at 0.4 rate and batch normalization layers to prevent overfitting constituted the model; categorical cross-entropy was used for calculating the loss. The learning rate was in the normal level even if it's at maximum speed if the input picture size is 224x224 pixels. The findings are such that the algorithm detects many hand motions exactly in the same manner.