Accurate human activity recognition with multi-task learning
- PDF / 2,001,749 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 91 Downloads / 303 Views
REGULAR PAPER
Accurate human activity recognition with multi‑task learning Yinggang Li1 · Shigeng Zhang1 · Bing Zhu1 · Weiping Wang1 Received: 14 June 2020 / Accepted: 16 September 2020 © China Computer Federation (CCF) 2020
Abstract Human activities recognition (HAR) in wearable devices is a promising technology in pervasive computing. However, the traditional method often regards human activity recognition as a single label recognition problem, ignoring the association between the current activity mode, personal motion mode and sensor wearing position. This paper proposes a multi-task human activity recognition multi-task learning framework based on supervised learning, which not only considers the activity, but also considers the identity of the wearer, gender and the position of the sensor on the body. We extracted the timedomain and frequency-domain features of the original data, and classified the data through a multi-task learning framework composed of a fully connected network and a convolutional neural network. We employ a public data set composed of 15 experimenters, 8 movements and 7 body positions. Only 30% of the data is used to train the model, which can achieve high precision. The experimental results show that the classification accuracy of activity recognition can reach 90.8% , body position recognition can reach 98.7% , wearer identity recognition can reach 97.5% , gender recognition can reach 98.7% . We call the model trained with 30% data as a pre-trained model, and then put personal data into the pre-trained model for fine-tune. Using a pre-trained model for fine-tune on personal data can achieve up to 95.6% activity recognition accuracy. Keywords Neural network · Machine learning · Human activity recognition · Pervasive computing
1 Introduction In the past few years, wearable devices have shown a blowout trend, and researchers are increasingly concerned about the application of universal wearable devices in human health. An important application of wearable devices is HAR (Wang et al. 2019). The wearable device can detect the user’s heart rate under low power consumption, identify the user’s actions, and apply this information to provide timely feedback on the user’s physical condition. These data are derived from the three-axis acceleration generated by the accelerometer and gyroscope placed in the device. Many * Shigeng Zhang [email protected] * Bing Zhu [email protected] Yinggang Li [email protected] Weiping Wang [email protected] 1
School of Computer Science and Engineering, Central South University, Changsha 410083, China
studies rely on the data from these two sensors to achieve good results on HAR. In recent years, there have been many studies on HAR, and most of these studies rely on the three-axis acceleration generated by accelerometers and gyroscopes placed in wearable devices (Chen and Xue 2015; Coskun et al. 2015; Kwapisz et al. 2010; Zhang et al. 2019). The most common method is to extract statistical features from the original activity data, and use machine lear
Data Loading...
