Zero shot vehicle extraction improves point cloud completion in underground parking
ZVeC: A Zero-Shot Framework for Instance-Level Vehicle Extraction and Generative Point Cloud Completion
Computer Vision and Pattern Recognition
Summary
3D scans of underground parking lots often miss parts of vehicles because of obstacles and limited viewpoints. The authors created ZVeC, a method that breaks down the scene into individual vehicles and completes each one separately without needing special training for each parking lot. They use a special model that understands general vehicle shapes to fill in missing parts, then put the scene back together. Tests show ZVeC works well even when given very sparse or incomplete input data.
What this means in practice
- •For autonomous vehicle teams: Improve 3D mapping accuracy in underground parking garages by completing partial LiDAR scans of vehicles without additional training data for each location.
- •For smart city planners: Generate complete 3D models of parked vehicles in underground garages from sparse LiDAR data to assist in space utilization and management.
Authors
Daisy Li, Kyle Gao, Quanyun Wu, Boris Jutzi, John S. Zelek, Jonathan Li
Abstract
LiDAR point clouds acquired in underground environments exhibit severe geometric incompleteness due to occlusions and limited sensor viewpoints, making reliable point cloud completion challenging without large supervised datasets. We propose ZVeC, a zero-shot, instance-driven framework that reformulates scene-level completion as compositional object-level reconstruction. By decomposing a scene into semantic object instances, ZVeC reduces reconstruction ambiguity in cluttered environments while eliminating the need for scenario-specific training. Each segmented vehicle is completed independently using a depth- and 3D Gaussian-conditioned diffusion model that exploits generalized geometric priors before the reconstructed instances are recomposed into the original scene. To evaluate our approach, we construct a real-world dense LiDAR benchmark of underground parking environments. Experimental results demonstrate consistent improvements over representative scene-level baselines in both quantitative metrics and visual quality. The completed point cloud differs substantially from the measured input (average KL divergence ~ 2.1), yet reducing the input to only 1% of the original LiDAR measurements changes the completed reconstruction only marginally (KL divergence < 0.50). This demonstrates that ZVeC produces geometrically consistent completions even under extreme input sparsity.