A Study of Hand-Raising Detection Method Using MediaPipe Framework and CNN Deep Learning Model

Abstract

The objectives of this research are: 1) to develop the steps of a hand-raising detection method with the MediaPipe framework and CNN deep learning model, and 2) to evaluate the performance of the steps of the developed method.

           The develop steps of hand-raising detection method that hand position above a head consist of 1) using input images that are 70 images of hand-raising and no hand-raising, 2) processing the input images with the MediaPipe framework for identifying the raised hand position and face position of raised hand person, 3) identifying the images of the hand and face positions use for training and testing the CNN deep learning model, and 4) identifying a predicted result of hand-raising above the head. The development of the proposed method uses the Python language, the MediaPipe framework library, the OpenCV library, and the TensorFlow library.

           The research results show that: 1) The developed method uses the MediaPipe framework for preprocessing and the CNN deep learning model for the prediction of hand-raising detection. The CNN structure has a dense two-layer hidden layer (64 and 32 units with the ReLU function) and one output layer with the Sigmoid function. 2) The evaluation results show a precision of 96.20% in detecting hand-raising above the head.