Hierarchical Multi-Classification for Sensor-based Badminton Activity Recognition

Fast development of sensor technology makes sensor equipments more and more smart and wearable. It further boost the need of sensor-based human activity recognition. Due to the lack of large-scale labeled datasets in practical AI applications, it is important to utilize prior information of the categories in sensor-based human activity recognition. In this paper, we propose a Hierarchical Multi-Classificaion (HMC) framework for sensor-based badminton activity recognition with the help of the prior information of badminton activity categories. Specifically, the multi-class sensor-based badminton activity recognition task is performed in two steps: (1). Any input data for a badminton activity are classified into one of the major classes which are based on their characteristic features; (2). They are further classified into one of the specific categories of badminton activity as required. It is demonstrated by the experimental results on BSS-V2 dataset that our proposed method can get up to 83.9% badminton activity recognition accuracy which is 1.7% better than previous state-of-the-art models.