The Future of Human Activity Recognition: Deep Learning or Feature Engineering?

A significant gap exists in our knowledge of how domain-specific feature extraction compares to unsupervised feature learning in the latent space of a deep neural network for a range of temporal applications including human activity recognition (HAR). This paper aims to address this gap specifically for fall detection and motion recognition using acceleration data. To ensure reproducibility, we use a publicly available dataset, UniMiB-SHAR, with a well-established history in the HAR literature. We methodically analyze the performance of 64 different combinations of (i) learning representations (in the form of raw temporal data or extracted features), (ii) traditional and modern classifiers with different topologies on (iii) both binary (fall detection) and multi-class (daily activities of living) datasets. We report and discuss our findings and conclude that while feature engineering may still be competitive for HAR, trainable front-ends of modern deep learning algorithms can benefit from raw temporal data especially in large quantities. In fact, this paper claims state-of-the-art where we significantly outperform the most recent literature on this dataset in both activity recognition (88.41% vs. 98.02%) and fall detection (98.71% vs. 99.82%) using raw temporal input.