Deep Learning Algorithms for Environment Perception

This thesis explores deep learning algorithms for environment perception, aimed at enhancing autonomous driving and smart-city infrastructure. Beginning with Vehicle-to-Everything (V2X) technology, it highlights the 5GMETA platform’s role in enabling real-time data for applications like parking detection and traffic safety. A novel method for aligning traffic camera images with satellite data improves object geo-location for V2X systems. The thesis also focuses on LiDAR-based perception for high-speed driving, presenting optimized models for 3D detection and effective domain adaptation techniques. Integrating V2X, deep learning, and computer vision, this work advances perception technologies for safer, smarter cities.