Novel measurement techniques for motion analysis and human recognition by using ultrasound and radiofrequency signals

Human detection and identification has been constituting a fundamental application in everyday life, and it is assuming an increasing importance both within the commercial world and the scientific community. The applications cover a variety of different areas, such as security and surveillance, rescue/recovery of troubled people, medicine and gait analysis. Among the plenty of employed technologies, over the last few years, radar and ultrasound systems are gaining an increasing interest in human detection/identification scenario. They can effectively work in particular environments and situations where other systems may fail, like in the darkness, in smoky or foggy areas, or through not transparent obstacles. Many radar and ultrasound systems for human detection exploit Doppler and micro-Doppler effects to analyse the target motion. The micro-Doppler effect is a variant of the Doppler phenomenon which accounts also for micro-motions of the target superimposed to the bulk motion. Hence, it can be deployed to record the micro-Doppler characteristics of different moving targets, commonly referred as micro-Doppler signature. Human moving targets exhibit very distinctive and unique micro-Doppler signatures, which, if properly analysed, may be used to carry out human target recognition and classification tasks. Aim of this dissertation is to provide a general and technology independent measurement approach suitable for human identification purposes. The approach is a list of elementary blocks covering the measurement device definition and setup, the micro-Doppler signature acquisition and the algorithm for signatures analysis. In order to test the procedure, several experimental trials to collect micro-Doppler signatures have been performed, both in the ultrasonic and radio frequency domain. A novel algorithm has been also designed and developed to extract some particular features from the acquired signatures, to be used for target classification. Recognition performance has been assessed as a function of some key algorithm parameters to investigate the level of robustness of the proposed features. Results show that high level recognition performance can be achieved for different human activities and subjects, both employing ultrasounds and radio-frequency waves