Jing Zhang, Meng Cheng Lau, Ziping Zhu

Abstract: We introduce a hybrid CNN-GRU model in this study to classify exercises using IMU time-series data, with a focus on jumping jacks, lunges, and squats. By combining Convolutional Neural Networks with Gated Recurrent Units, our model effectively manages the high dimensionality and variable sampling rates of IMU data. We employed data normalisation and augmentation techniques to refine the dataset. Our model showed high accuracy in classifying types of exercises, highlighting its potential in motion classification and fitness-tracking applications. These results emphasise the value of hybrid deep learning methods in analysing complex time-series data and make a significant contribution to the understanding of human exercise movement patterns.

Keywords: Motion Classification, IMU, Deep Learning, CNN-GRU.

Date Published: September 6, 2024
DOI: 10.11159/jmids.2024.007

View Article