Multi-target human gait classification using LSTM recurrent neural networks applied to micro-Doppler

A Recurrent Neural Network (RNN) with a Long Short Term Memory (LSTM) implementation can be effectively utilized to classiiy the radar signature of zero to three human gaits, based on their micro-Doppler return. The LSTM-RNN outperforms the current state-of-the-art with an accuracy of 89.1%. The use of a sequence-to-sequence classification method like RNN additionally brings the benefit of being able to classify measurements with a variable observation time which, on average, is notably less than the state-of-the-art. Another advantage of the LSTM-RNN is that it can also be used with measurements that include transitions between classes over time.