Deep learning for sensor-based activity recognition: recent trends

Abstract

The field of human activity recognition (HAR) using different sensor modalities poses numerous challenges to the researchers working in this domain. Though traditional pattern recognition approaches performed well in this regard earlier, the cost of poor generalization and the cost of shallow learning due to the handcrafted features have opened a new door for deep learning in this field. This chapter discusses the importance of deep learning in sensor-based activity recognition explaining the deep models and their use in previous research works. This chapter also represents the importance of transfer learning and active learning in this field, that are new research topics. Finally, this chapter shows the challenges of using deep models along with feasible solutions.