Deep Learning Ensemble for Recognising Lower Limb Activity

Security, healthcare, elderly care, rehabilitation, and sports science are just a few of the areas that can benefit from the analysis of lower limb motion and human activity recognition (HAR). In order to improve the accuracy of the HAR system, a novel deep learning ensemble (DL-Ens) model composed of three lightweight convolutional and recurrent neural networks is presented in this study. Evaluation of the activity recognition performance of the suggested DL-Ens approach is carried out on a self-recorded dataset acquired using multiple wearable motion sensors as well as on the publicly accessible UCI's human activity recognition (UCI-HAR) dataset. The individual deep learning models are tested for time-series classification. However, the proposed DL-Ens approach achieves the highest classification accuracy of 97.48 +/- 5.02% on the self-recorded dataset and 93.36 +/- 5.89% on the UCI-HAR dataset.