Detection of Multiple Humans Equidistant From IR-UWB SISO Radar Using Machine Learning

In order to locate humans unconscious under rubble, we generally use an ultrawideband impulse radar to capture the chest displacement caused by breathing. Sometimes, cases may arise when more than one subject falls in the field of view of the radar antenna and are located equidistant from it. At that time, it becomes impossible for the radar to count the actual number of people present. Thus, we created a scenario under laboratory conditions where we placed three subjects equidistant from the radar with no obstruction in between. Then, we applied supervised machine learning algorithms to classify and predict the number of subjects using variance and singular values as the features. We found that the ensemble bagged trees classifier gave the most accurate classification at 97.5%.