一种基于视觉大模型的稀疏点云引导视频深度预测方法
By using visual large model-guided depth completion and high-resolution depth inference, combined with sparse point clouds and visible light video, the problems of temporal instability and lack of scale information in depth prediction are solved, and high-precision and stable video depth generation is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG UNIV
- Filing Date
- 2025-08-13
- Publication Date
- 2026-07-17
AI Technical Summary
Existing depth estimation methods based on large visual models cannot handle sparse point clouds and visible light videos, resulting in a lack of temporal stability and accurate scale information in depth prediction, which limits their application in fields such as autonomous driving and 3D reconstruction.
By using visual large model-guided depth completion and high-resolution depth inference, combined with sparse point clouds and visible light video, and utilizing the local least squares method of spatiotemporal neighborhood and the temporal alignment module, temporally stable high-resolution video depth is generated.
It improves the accuracy and temporal stability of depth prediction, and can extract robust scene features from sparse point clouds and visible light videos to generate detailed high-resolution video depth.
Smart Images
Figure CN120997272B_ABST