Deep Learning in Indoor Human Activity Recognition with Millimeter Wave Radar
Human activity recognition (HAR) is an active field of study concerned with automatically identifying human activities by analyzing and classifying the data captured from sensors. HAR has received great attention due to its wide application in healthcare and eldercare. Deep learning models such as multilayer perceptron (MLP), convolutional neural network (CNN), and recurrent neural network (RNN) have been widely applied to HAR. Recently, Transformer emerges as a new type of deep learning neural network, which has been a breakthrough in natural language processing tasks, and is showing great promise in the field of computer vision. In our research, we experimented with three Transformer models and three non-transformer models using a public indoor millimeter wave radar dataset called MMActivity. The highest prediction accuracy of 96.10% was achieved by a Transformer model integrating CNN with the Transformer encoder, which is comparable to the prediction accuracy of 95.90% achieved by the non-transformer model BiLSTM. When applying to the dataset collected by our own millimeter wave radar, our Transformer model integrated with CNN also showed a high prediction accuracy of 93.6%.
Committee: Carl Chang (major professor), Simanta Mitra, and Ying Cai
Meeting number: 2620 393 9453 Password: a2YJS55cxej