Autonomous driving paper index
HESS: Height-Aware Benchmarking of Monocular Depth Estimation Using Synthetic and Real Datasets
One-line summary
We propose a height-aware evaluation framework for monocular depth estimation (MDE) that leverages the MultiHeightView dataset, combining real and synthetic traffic scenes across multiple viewpoints.
Engineering notes
Accurate depth estimation is vital for autonomous driving and traffic monitoring, yet benchmarking remains difficult due to the scarcity of LiDAR ground truth.
Chinese explanation / 中文解读
中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。
Original abstract
Accurate depth estimation is vital for autonomous driving and traffic monitoring, yet benchmarking remains difficult due to the scarcity of LiDAR ground truth. We propose a height-aware evaluation framework for monocular depth estimation (MDE) that leverages the MultiHeightView dataset, combining real and synthetic traffic scenes across multiple viewpoints. The framework introduces the Height Error Sensitivity Score (HESS), a diagnostic metric that quantifies robustness to height variation. Pseudo-ground-truth depth is validated using manual laser measurements and an Intel RealSense sensor, offering a cost-effective alternative to LiDAR. We also implement a mask-guided fusion strategy that integrates Depth Pro with YOLOv11 segmentation, enabling object-specific analysis of near-field accuracy versus cross-height robustness. Experiments show that Depth Pro struggles with occlusion, far-field objects, and viewpoint shifts, while fusion improves near-field accuracy but reduces stability across camera heights. To confirm that height sensitivity is not model-specific, we include MiDaS as a qualitative baseline, which exhibits similar trends across heights.
Links and sources
Need this topic turned into a technical roadmap?
Full Self Driving can prepare a custom autonomous driving literature review, code map, dataset map, and B2B technology assessment.
Request B2B research
Comments