As Autonomous Vehicles advance toward higher levels of automation, real-time perception and accurate mapping must converge to enable safe and scalable deployment. This session explores how modern computer vision techniques are powering next-generation HD maps, helping vehicles interpret and adapt to changing urban conditions. From on-the-fly map updates to understanding scene semantics, we’ll examine how vision systems are becoming core to dynamic localization and navigation.

• How computer vision pipelines enable continuous map updates through crowd-sourced perception
• Leveraging semantic segmentation and object tracking for map-layer enhancements
• Addressing occlusions, shadows, and unpredictable actors in urban settings
• Lessons from real-world deployment: vision-based map updates vs. traditional HD maps