Answers are grounded in this paper and cite their sources.
Using a single 3-axis accelerometer for human activity recognition is challenging but attractive in daily healthcare and monitoring with wearable sensors and devices. In this paper, an effective and efficient framework is proposed to address the recognition problem without any heavy preprocessing. We encode 3-axis signals as 3-channel images using a modified recurrence plot (RP) and train a tiny residual neural network to do image classification. The modified RP is first proposed in our paper to overcome its tendency confusion problem, which has improved our system performance significantly. We evaluate our algorithm on a new database and a public dataset. Results show that our recognition framework achieves highly competitive accuracies and good efficiencies with other state-of-the-art methods on both datasets. Moreover, our method shows stronger robustness to noise and low decimation rate through comparison experiments. Finally, we provide detailed discussion and analysis of our approach from two perspectives: the pattern analysis of encoding algorithm and the interpretation of classification model.
No takes yet. Share an insight, caveat, or question.
Lu et al. (2019) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: