This thesis project focuses on biometric recognition based on human gait analysis. During an experimental measurement campaign, radar signals were acquired from moving subjects. These radar data allow for the extraction of not only the user's position but also the micro-Doppler components—characteristic micro-movements, such as pelvic oscillations or step cadence, which serve as a unique "fingerprint" of an individual's gait.

In parallel with the radar measurements, data were recorded from the Inertial Measurement Unit (IMU) of a smartphone held by the subject. The objective of this thesis is to investigate the correlation between the features extracted from the radar and those detected by the accelerometer for identification purposes.

Specifically, the study will focus on the following key areas:

• IMU-based Identification: Evaluating the effectiveness of inertial information (tri-axial acceleration) in discriminating and identifying users through classification models.

• Cross-Modal Signal Synthesis: Assessing the feasibility of generating synthetic IMU signals starting from radar data. The goal is to reconstruct the dynamic movement characteristics observed remotely by the radar and project them into the domain of the inertial sensor.

• Vulnerability Analysis (Spoofing): Demonstrating how a "fake signal" generated from radar data can deceive an IMU-based authentication system, enabling the identification of a subject through a synthetic signal that was never actually recorded by the physical device.

The project involves the development of algorithms in Python, utilizing specialized libraries for signal processing and the implementation of Machine Learning models.