Improved IMU-based Human Activity Recognition using Hierarchical HMM Dissimilarity

Although there are many classification approaches in IMU-based Human Activity Recognition, they are in general not explicitly designed to consider the particular nature of human actions. These actions may be extremely complex and subtle and the performance of such approaches may degrade significantly in such scenarios. However, techniques like Hidden Markov Models (HMMs) have shown promising performance on this task, due to their ability to model the dynamics of such activities. In this work, we propose a novel classification technique for human activity recognition. Our technique involves the use of HMMs to characterize samples and subsequent classification based on the dissimilarity between HMMs generated from unseen samples and previously-generated HMMs from training/template samples. We apply our method to two publicly-available activity recognition datasets and also compare it against an extant approach utilizing feature extraction and another technique utilizing a deep Long Short-Term Memory (LSTM) classifier. Our experimental results indicate that our proposed method outperforms both of these baselines in terms of several standard metrics.