Pose Estimation of Player's Hand with CNN for the Hand Pose Rally System
Motoya Hirakawa, Akira Suganuma · Institutional Repositories DataBase (IRDB) · 2018
Study on hand pause rally system has been taking place since 2012 in our laboratory. In this study, a method identifying the posture of a human hands has been developed. The conventional method has several problems. One of them is that accuracy does not readily rise. Therefore, it is necessary for us to build a method of identifying the posture by machine learning using a convolution neural network (CNN) which is an alternative method to the conventional discrimination method. As a result of the verification, the accuracy of the identification was about 40% overall in spite of a small learning data. Therefore, we processed the image data and artificially increased the amount of images. As a result of increasing the learning data and measuring it, the accuracy reached 54%. Currently we are studying ways to improve accuracy other than increasing the amount of images.