LSTM based Human Activity Classification on Radar Range Profile

Abstract

A bi-directional long short term memory(LSTM) based deep learning approach to classify human activities with radar high resolution range profiles(HRRPs) is investigated. MOCAP dataset, from Carnegie Mellon University, is used for HRRPs simulation. Six activities are classified with the proposed network and an appreciable classification result has been acquired. Experiment demonstrates that bi-directional LSTM performs better than unidirectional LSTM in this study. We also exam the activity duration of every piece of data to find out its impact on classification performance.