一种结合深度学习模型与几何模型的视觉定位方法及系统
By combining deep learning models with geometric models in visual localization, an information field is constructed and spatial coordinates are optimized, solving the problems of feature point loss and mismatch in complex environments in traditional visual localization methods, and achieving high-precision visual localization results.
Patent Information
- Application Number
- CN202610492847.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-17
- Estimated Expiration
- 2046-04-15
AI Technical Summary
Traditional visual localization methods are prone to feature point loss or mismatch in scenes with weak texture, repetitive texture, drastic lighting changes, or motion blur. Furthermore, deep learning methods have limited generalization capabilities, making it difficult to meet the requirements of high-precision applications. Sparse point cloud data cannot encode the geometric density and semantic category of a scene, causing SLAM systems to fail in localization in dynamic environments.
By combining deep learning models and geometric models, an information field is constructed that simultaneously encodes scene geometric density, semantic category probability, and local feature descriptors. Through feature extraction and motion analysis, the pose of the next frame is predicted, and spatial coordinates are optimized to improve positioning accuracy.
It improves the accuracy and robustness of visual positioning, enabling it to handle dynamic objects in complex environments and enhancing positioning stability and accuracy in dynamic environments.
Smart Images
Figure CN122023755B_ABST
Abstract
Citation Information
Patent Citations
Scene space three-dimensional model dynamic modeling method based on multi-modal data
CN119339008A
KR20220081261A