Towards the Detection of Violent Scenes at Public Places by Analyzing Range-Doppler Maps

Cameras are usually applied for the surveillance of public places. We investigate the use of radar technology as substitute for the camera-based supervision of public places where a high potential for the occurrence of violent conflicts exists. The goal is the setup of a surveillance system that respects the privacy of people to a remarkable higher extent than camera-based systems. This will make it also easier for police units to legally observe such public places. We present the technical setup of a radar module that determines range-Doppler maps as output. The time sequence of these maps is taken as input to detect and classify dangerous and violent scenarios with a classification system based on the application of neural networks. Micro-Doppler spectra captured in a first test phase are presented for different scenarios where people are in danger or violence could occur. We give an outlook how the classification system and the structure of the neural network will be designed.