A Decade of Progress in Human Motion Recognition: A Comprehensive Survey From 2010 to 2020

Abstract

The central perspective of this review is to categorize research in Human Motion Recognition (HMR) over the past decade into two significant categories: vision sensor-based (VS) methods and wearable sensor-based (WS) methods. Within each category, research is further assessed from the viewpoints of sensors, classification algorithms, datasets, gesture types, target body parts, and performance. This approach allows for a comprehensive assessment of the overall research trends and technological advancements in HMR. Both VS methods and WS methods present their own sets of advantages and challenges. VS methods face challenges related to limited workspace, varying lighting conditions, occlusion, and complex image processing. Conversely, WS methods, compared to VS methods, deals with challenges associated with multiple sensor calibration, intrusiveness, and magnetic field mapping due to sensor placement. As such, the choice between these methods depends on the specific application, the required level of accuracy, and user preferences. Gaining insights into the nature of various HMR methods and staying informed about recent research trends is of utmost importance. By the end of this review, readers will gain a comprehensive and systematic understanding of the latest developments in HMR techniques, which will serve as a valuable resource for researchers and practitioners alike.

Keywords

• Human gesture recognition
• human motion recognition
• human-robot interaction
• wearable sensor
• vision sensor