AI-based Rock-Paper-Scissors plug and play system

Abstract

In this paper, we present a plug and play Rock-Paper-Scissors game set that utilizes CNN-based static hand gesture recognition and a Markov chain-based bot. Each game round, the camera output is fed to the neural network and the hand gesture result is compared to the current Markov chain state. We utilize a CNN trained with a dataset containing rock-paperscissors hand gestures and unwanted inputs, and a multi-state Markov chain. Automated tests confirm the bot prediction capabilities and practical tests verify the precision of static hand gesture recognition. Tests against a verification dataset reveal a recognition accuracy of 97.4%.