Human Behavior Recognition Using Micro-Doppler Signatures and 2D-CNN-LSTM Network

2-dimensional convolutional neural network and long short-term memory (2D-CNN-LSTM) network and micro-Doppler signatures were used to recognize human behavior in this paper. First, frequency modulated continuous wave (FMCW) radar was used to acquire raw radar data. Then, the collected data were preprocessed to produce micro-Doppler signatures. The micro-Doppler signatures have temporal and spatial information. Hence, the 2D-CNN-LSTM network is proposed. The section of convolutional neural network (CNN) layers and a convolution long-short term memory one-dimension (ConvLSTM1D) layer were used to tackle temporal and spatial information. The long-short term memory (LSTM) layer was used to tackle temporal information. The 2D-CNN-LSTM network was experimentally found to be lightweight while achieving high accuracy.