A study on dynamic hand gesture recognition for finger disability using multi-layer neural network
Romi Fadillah Rahmat, Sarah Purnamawati, Eka Pratiwi Goenfi, Opim Salim Sitompul, Muhammad Fermi Pasha, Rahmat Budiarto · Monash University Research Portal (Monash University) · 2018
Interaction between human and computer is generally performed with a keyboard and mouse. However, these interactions have certain drawbacks which cannot be done by users with physical disabilities or user who have disability from the wrist to the fingertip. To overcome this problem, an approach to recognize human hand gesture as a means of human-computer interaction is needed. The method proposed by the author is the use of algorithms: nearest neighbor, grayscaling, frame-differencing, Principal Component Analysis (PCA) and Multi-Layer Perceptron (MLP). This research was conducted in two experiments, which were experiment with six different types of hand gestures and experiments with four different types of hand gestures. Each experiment was performed five times with different value of number of hidden layers parameter and hidden neurons parameter. The best testing result obtained from the experiment with six types of hand gestures is from the second experiment with two hidden layers using 300 and 50 hidden neurons for each layer, resulting in an accuracy rate of 77.02%. The best testing result obtained from the experiment with four different types of hand gestures is from the first experiment with two hidden layers using 300 and 50 hidden neurons for each layer, resulting in an accuracy rate of 89.72%. The best overall result is then implemented into the front-end system for controlling application such as: file explorer, music player, video player, slideshows and PDF reader.