Advanced 3D perception based on 2D projection and sensor fusion for autonomous vehicles

Deep learning (DL) has experienced rapid growth in recent years due to its universal learning approach, robustness, generalization, and scalability. Furthermore, the field of autonomous driving cars (ADC) is continuously advancing, with the prospect of fully automated public vehicles becoming a near...

Descripción completa

Detalles Bibliográficos
Autor: Guerrero Casado, Quim
Tipo de recurso: tesis de maestría
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/404036
Acceso en línea:https://hdl.handle.net/2117/404036
Access Level:acceso abierto
Palabra clave:Automated vehicles
Artificial intelligence
LiDAR
Artificial Intelligence
object detection
perception
autonomous vehicles
Light Detection And Ranging
Vehicles autònoms
Intel·ligència artificial
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Telecomunicació òptica::Fotònica
Descripción
Sumario:Deep learning (DL) has experienced rapid growth in recent years due to its universal learning approach, robustness, generalization, and scalability. Furthermore, the field of autonomous driving cars (ADC) is continuously advancing, with the prospect of fully automated public vehicles becoming a near-future reality. In this context, a critical safety aspect for ADCs revolves around the precise recognition of pedestrians and obstacles. This thesis aims to tackle this challenge by using 2D images obtained from the projections of the 3D LiDAR data. With those 2D images, we will be able to use fully developed 2D neural networks instead of 3D neural networks that are less mature. We compared our method and results with the results obtained using only RGB raw images to check its effectiveness, but using the RGB images led to better outcomes.