A Real-Time Lightweight Method to Detect the Sixteen Brushing Regions Based on a 9-axis Inertial Sensor and Random Forest Classifier

Abstract

In this research, we propose a real-time lightweight method to detect the tooth-brushing regions. The system takes a sensor node as input device and a 2D dentition as output interface. Specifically, the sensor node can be attached onto the handle of a toothbrush to get the 3D Euler angles, which are used as the features for training Random Forester Classifier (RFC) Model. Moreover, the predicted brushing region will be displayed on the dentition interface in real-time. All teeth are divided into 16 regions to evaluate the detection accuracy. User-dependent offline experiment shows that the RFC-based automatic threshold definition method achieves 97.6% validation accuracy, which is about 35% higher than the manual threshold definition method. For the RFC-based method, the online real-time accuracy is 74.0%, which is about 23.6% less than the offline result.