Radar-Based Multiple Target Classification in Complex Environments Using 1D-CNN Models

In this paper, we propose a robust multiple target classification algorithm for real-world complex cluttered environments that can be mapped into low-cost millimeter-wave (mmWave) sensors considering limited memory and processing power budget. A novel approach is developed to create both mu-Doppler and mu-range spectrogram of multiple objects concurrently using an extended Kalman filter (EKF) based tracking layer integration. One-dimensional (1D) time sequence features are extracted from both spectrograms per target object, and a 1D convolutional neural network (CNN) based classifier is built to classify multiple target objects (human or non-human) in the same scene accurately.