Using additional training sensors to improve single-sensor complex activity recognition
Paula Lago, Moe Matsuki, Kohei Adachi, Sozo Inoue · 2021
We propose a method for single-sensor based activity recognition using multiple sensors during training time. The proposed method, based on learning a shared representation space, can be used to improve the accuracy and F-score of complex activity recognition with a single on-body accelerometer sensor by leveraging data from other sensors at training time. Results show improvements of 16% in accuracy and 20% in F-score.