Carranza García, ManuelGarcía Gutiérrez, JorgeRiquelme Santos, José Cristóbal2024-12-202024-12-202021-09Carranza García, M., García Gutiérrez, J. y Riquelme Santos, J.C. (2021). Artificial intelligence techniques for improving the perception systems of autonomous vehicles. En XIX Conferencia de la Asociación Española para la Inteligencia Artificial. CAEPIA 20/21, Málaga, España.https://hdl.handle.net/11441/166030Autonomous driving has the potential to revolutionize mobility, but still presents many challenges that need to be addressed. The correct perception of the environment using remote sensing data is one of the fundamental aspects in which improvements are required. In this thesis project, our aim is to enhance the perception systems of autonomous vehicles, addressing the object detection task from different perspectives. Recent works in 2D object detection are mainly focused on the general-purpose COCO benchmark. However, their use in other domains such as autonomous driving, in which real-time inference is required, is yet an area to be fully explored. After performing an experimental revision of the speed/accuracy tradeoff of existing deep learning-based 2D object detectors, our goal is to design a detection system that is adapted for this scenario. The proposed detector considers aspects such as the perspective projection of the vehicle’s cameras and the high imbalance between different instance classes, while also meeting the efficiency requirements. Furthermore, we aim to develop temporal methods that incorporate information from past frames to improve detection precision. Then, the last objective is to increase the robustness of the detector by exploring data fusion techniques between 2D camera and 3D LiDAR data. The main data resource for the project will be the Waymo Open Dataset, which is the most extensive and diverse multi-modal self-driving dataset up to date.application/pdf6 p.engAttribution-NonCommercial-NoDerivatives 4.0 Internationalhttp://creativecommons.org/licenses/by-nc-nd/4.0/Autonomous drivingConvolutional neural networksData fusionDeep learningObject detectionPerceptioninfo:eu-repo/semantics/conferenceObjectinfo:eu-repo/semantics/openAccess