Depth Vision Guided Human Activity Recognition in Surgical Procedure using Wearable Multisensor

The increasing complexity of modern surgery rooms brings many challenges. Human activity recognition (HAR) plays a significant part in healthcare, telemedicine, long-term treatment, and even surgery by using wearable inertial sensors or depth cameras. Although the development of artificial intelligence techniques provide various machine learning (ML) methods to identify activities, it is a time-consuming implementation and high work burden to collect and label the large data set. To fascinate efficient data collection and labeling, we propose a novel depth vision-guided HAR architecture to obtain the labels of the collected raw data from the inertial measurement unit (IMU) sensors using depth data automatically. Experimental results show that the novel depth vision-guided interface can be utilized for identifying activities without labeling the data in advance.