MORE | Spring 2025

Optimizing Perception Capabilities of Autonomous Vehicles Through V2I Late Fusion using Kalman Filter

Perception is a critical component of Autonomous Vehicles (AVs). However, onboard sensors frequently encounter limitations such as occlusions and blind spots caused by surrounding vehicles, infrastructure, or environmental obstacles. Vehicle-to-Infrastructure (V2I) communication integrates real-time data from roadside units (RSUs) to enhance AV perception capabilities. This research presents a V2I framework that uses monocular traffic cameras installed at intersections for 3D object detection. The framework combines roadside perception data with onboard LiDAR sensor information through a late fusion technique utilizing a Kalman filter for synchronization and refinement. Simulation scenarios involving occlusion at intersections are reconstructed in CARLA to evaluate the proposed approach in terms of perception range and accuracy.