Object Classification With Micro-Doppler Spectra For Surveillance Applications

The Micro-Doppler spectrogram of a 24 GHz radar module is used to classify objects in a surveillance scenario. Image data are synchronously captured to support the annotation of collected radar data by an additional image recognition. A neural network is trained and applied to assign an unknown object to a limited number of classes with the intention to detect moving persons in the scenario and distinguish them from other classes like vehicles. Recognition can be improved by using range and azimuth angle as additional features for the classification that supports the learning of different spectrogram characteristics dependent on distance and moving direction of an object.