Gesture Recognition and Showing a Cyclist's Intent

Thomas Bridgeman, Helena Gibson, Kevin Meehan

Abstract: Gesture recognition is an important step forward for both humans and machines but especially the interaction between them. The transport industry is a sector that can benefit greatly due to the reliance on using one’s hands to steer a car or cycle a bike. Bicycles often prove to be challenging regarding showing the intent of the cyclist to drivers, other cyclists and pedestrians. This paper showcases the development of a gesture recognition system that translates a cyclist’s gestures to better show their intentions to other road users through the use of an Arduino and LED. The solution uses the MediaPipe framework and a Convolutional Neural Network. It was trained and tested on using a custom dataset. There were four different classes of gestures (neutral, stop, direction and thanks) and provides a high degree of accuracy in recognition.