Detection of Violent Scenes at Public Places by Classifying Range–Doppler Maps

A classification system for analyzing sequences of range-Doppler maps with neural networks is presented. The system aims at the detection and classification of dangerous or violent scenes in public places. We have installed radar modules inside and outside a train station to capture radar data in these scenarios. Recognition experiments are performed on these data. Augmentation methods have been developed to artificially create additional training data. Recognition results are presented.