A Review of Wearable Sensor-based Human Activity Recognition using Deep Learning

Human activity recognition is an important direction in pattern recognition that learns from low-level raw signals acquired from smartphones and commercially available and customizable wearable devices to acquire high-level knowledge. HAR plays an essential role in providing smart healthcare to physically impaired older adults, with potential applications for elderly care, fall detection physical rehabilitation, clinical assessment and surveillance. Numerous researchers and scholars have conducted HAR based on conventional pattern recognition (PR) approaches and deep models. Conventional PR methods rely on the heuristic hand-crafted feature, which needs to pre-process the raw signals. Deep learning models can automatically learn features end-to-end, compared with the conventional PR approaches have achieved promising performance. Therefore, this paper reviews the progress of activity recognition based on wearable sensor devices, and discussed the potential application areas of human motion recognition technology. Finally, this paper discussed the related problems that can be further studied in the field of activity recognition.