Abstract
In this paper, we propose an accurate and robust perception module forAutonomous Vehicles (AVs) for drivable space extraction. Perception is crucialin autonomous driving, where many deep learning-based methods, while accurateon benchmark datasets, fail to generalize effectively, especially in diverseand unpredictable environments. Our work introduces a robust easy-to-generalizeperception module that leverages LiDAR, camera, and HD map data fusion todeliver a safe and reliable drivable space in all weather conditions. Wepresent an adaptive ground removal and curb detection method integrated with HDmap data for enhanced obstacle detection reliability. Additionally, we proposean adaptive DBSCAN clustering algorithm optimized for precipitation noise, anda cost-effective LiDAR-camera frustum association that is resilient tocalibration discrepancies. Our comprehensive drivable space representationincorporates all perception data, ensuring compatibility with vehicledimensions and road regulations. This approach not only improves generalizationand efficiency, but also significantly enhances safety in autonomous vehicleoperations. Our approach is tested on a real dataset and its reliability isverified during the daily (including harsh snowy weather) operation of ourautonomous shuttle, WATonoBus