Human activity recognition with HMM-DNN model

Activity recognition commonly made use of hidden Markov models (HMMs) to exploit temporal dependencies between activities. The emission distribution of HMMs could be represented by generative models, such as Gaussian mixture models (GMMs), or discriminative models, such as random forest (RF). These models, especially discriminative ones, needed to manually extract features from the sensor data, which relied on the experience of the researchers, and usually was a time-consuming task when complicated features are extracted. Furthermore, with these methods, the process of quantization of the sensor data, i.e., manual feature extraction, might lose much useful information and thus led to a performance debasement. In this paper, we recommend deep neural networks (DNNs) for modeling the emission distribution of HMMs, which automatically learn features suitable for classification from the raw sensor data and then estimate the posterior probabilities of the HMM states. We collected a dataset of daily activities and based on which experiments were performed to compare our HMM-DNN model with both HMM-GMM and HMM-RF. The results illustrated that HMM-DNN outperformed both HMM-GMM and HMM-RF.