Classification models for data fusion in human activity recognition
Julien Maítre, Kévin Bouchard, Sébastien Gaboury · 2020
In this paper, we present alternative deep learning architectures that perform data fusion, more specifically, feature-level fusion in the context of human activity recognition. The proposed architectures combine statistical features from the time-domain and features extracted automatically with stacked convolutional layers. The power of these architectures relies on the fact that they can be fed by various sources of data (e.g., time series, images). Additionally, we exploited the publicly available Mobile Health dataset to assess the performances of the proposed architectures. The results obtained show that the architectures are suitable for future use.