Real-time Human Activity Classification From Radar With CNN-LSTM Network

Millimeter-wave radar is becoming an attractive solution to human activity classification for smart home monitoring, with radar's robustness and privacy-less advantage. Especially, the low signal-noise-ratio and real time requirement are meaningful to do further study on. In this paper, we propose a two-stream framework to detect human activity and classify the activity sequence simultaneously. We also fuse the range, velocity and spatial angle of target to improve the accuracy of classification. We gathered a human activity data set, containing 6 activities over 6 persons. The result of experiment shows that the detection-classification system achieves an average accuracy higher than 90% for classifier within 1 second, respectively.