Convolutional Long Short-Term Memory Networks for Doppler-Radar Based Target Classification

This paper investigates the use of range and Doppler features for target classification in Advanced Driver-Assistance Systems (ADAS). Pedestrians and bicyclists dataset was recorded using a 77 GHz Frequency Modulated Continuous Wave (FMCW) RAdio Detection And Ranging (RADAR) sensor. A Convolutional Long Short-Term Memory (CLSTM) based deep-learning model was proposed and evaluated with the recorded dataset. The proposed model uses a novel convolutional layer based feature compression method. Our proposed model was shown to generalize well to the dataset and was shown to perform with an average accuracy of 94.76% over the test subset.