An Online Adaptive Sampling Rate Learning Framework for Sensor-Based Human Activity Recognition

Abstract

In the field of sensor based human activity recognition, fixed sampling rate scheme is difficult to accommodate the dynamic characteristics of streaming data. It may directly leads to high energy consumption or activities detail missing problems. In this paper, an efficiency online activity recognition framework is proposed by integrating sampling rate optimization with novel class detection and recurring class detection algorithms. Based on the proposed framework, we believe that this system can effectively save battery life and computation capacity without decreasing the overall recognition performance.