Apparatus, device and method for performing a perception task, and corresponding computer-readable storage medium
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-08-19
- Publication Date
- 2026-03-26
AI Technical Summary
Existing computer vision techniques for autonomous driving suffer from inaccuracies in object detection and localization due to implicit height modeling in birds-eye-view (BEV) representations, particularly in terrains with varying elevations, leading to challenges in accurately detecting and localizing objects like traffic lights and signs.
Incorporating explicit height modeling alongside BEV space features by integrating explicit heightmaps with BEV representations, using sensors like LiDAR and cameras to capture precise object dimensions and positions, enhancing depth perception and adapting to varying terrains.
Improves the accuracy of object detection and localization in autonomous driving systems by providing additional depth cues, enabling more precise estimation of object distances, positions, and dimensions, especially in scenarios with occluded or partially obscured objects.
Smart Images

Figure US2025042540_26032026_PF_FP_ABST