Machine learning methods for radar-based people detection

• Autor(a) principal: Castanheira, José Pedro Baião
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/30403
• topic: People detection; Indoor environments; Radar; Machine learning; Mobile robotics

The present dissertation describes the development and implementation of a radar-based system with the purpose of being able to detect people amidst other objects that are moving in an indoor scenario. The detection methods implemented exploit radar data that is processed by a system that includes the data acquisition, the pre-processing of the data, the feature extraction, and the application of these data to machine learning models specifically designed to attain the objective of target classification. Beyond the basic theoretical research necessary for its sucessful development, the work contamplates an important component of software development and experimental tests. Among others, the following topics were covered in this dissertation: the study of radar working principles and hardware; radar signal processing; techniques of clutter removal, feature exctraction, and data clustering applied to radar signals; implementation and hyperparameter tuning of machine learning classification systems; study of multi-target detection and tracking methods. The people detection application was tested in different indoor scenarios that include a static radar and a radar dynamically deployed by a mobile robot. This application can be executed in real time and perform multiple target detection and classification using basic clustering and tracking algorithms. A study of the effects of the detection of multiple targets in the performance of the application, as well as an assessment of the efficiency of the different classification methods is presented. The envisaged applications of the proposed detection system include intrusion detection in indoor environments and acquisition of anonymized data for people tracking and counting in public spaces such as hospitals and schools.