Gesture Recognition by Machine Learning Combined with Geometric Calculation

Human Machine Interface (HMI) in the vehicle environment is becoming more complex due to an increased number of desired user functions. These added functions increase the driver's distraction since the driver must visually identify a particular physical interface that they want to control. A gesture recognition system is one possible method for controlling HMI functionalities without visual cues. Some technical approaches to gesture recognition that exploit deep learning have been proposed. However, these methods require increased computational cost in relation to processing time. Ideally, a gesture recognition system should enable fast and accurate driver action to avoid a distracted driver and maximize safety. In this study, we propose an algorithm that combines geometric calculation and machine learning techniques to estimate gestures with low computational overhead. The algorithm is explained in detail, and results are reported including a gesture recognition accuracy of 85% across three gestures for a moving hand from eight people.