一种物体级语义建图与工作站对齐方法及存储介质
By using RGB-D data and three-channel CLIP visual feature fusion in a chemical laboratory, combined with multi-view feature aggregation and three-level pose correction, the problems of mismatch, dilation and pose drift in object-level semantic mapping in chemical laboratories are solved, achieving stable object recognition and semantic embedding, supporting robot operations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- UNIV OF SCI & TECH OF CHINA
- Filing Date
- 2026-06-18
- Publication Date
- 2026-07-17
AI Technical Summary
Existing object-level semantic mapping methods suffer from problems such as misassociation of similar equipment, feature attenuation, point cloud expansion, and pose drift in chemical laboratory scenarios, making it difficult to achieve stable object recognition and semantic embedding.
RGB-D data is used for pose estimation. Combined with adaptive fusion of three CLIP visual features and multi-view feature aggregation, greedy allocation and dual anti-expansion gating are performed through joint scoring of geometric and semantic similarity. Combined with three-level pose correction and prototype verification, object-level mapping and workstation alignment are achieved.
It reduces mismatches with similar experimental instruments, avoids feature effect decay, controls abnormal expansion of point cloud size, corrects object map and pose deviations, completes occlusion and missed detection objects, and supports actual robot operations.
Smart Images
Figure CN122415700A_ABST