Recognition and Change Point Detection of Dogs' Activities of Daily Living Using Wearable Devices
Rina Amano, Jianhua Ma · 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech) · 2021
The purpose of this study is to monitor dogs' daily activities using wearable devices for capturing their movement data, which are utilized to detect the change points and recognize the activities. Two small and light devices are attached to the dog's collar and back to collect acceleration data and gyro data of Activity of Daily Living (ADL), including sitting, standing, lie down, walking, and running. In this research, we recognize ADL and detect change points using multiple wearable devices for evaluation of recognition performance in terms of devices' locations and combinations. The classifier is constructed by a deep learning algorithm of convolutional neural network (CNN), and the differences between the accuracy of the window size, sliding window size, the type of data, and the mounting position of the device were compared. As a result, when the window size was 3000ms, the average of five activity classifications is 92.6%, which is higher than the previous research using LDA and QDA. In the change point detection, we achieve an F1-score of 80.3% when the sliding window size is 900ms. It was also found that the performance in both activity classification and change point detection is more accurate when using both the acceleration data and the gyro data as compared with only using one of them. Further, the accuracy of the device is higher when both the back and the neck were used.