基于边界向量拟合的机器人多目标识别与抓取方法、系统、存储介质及计算机设备
By using boundary vector fitting, a U-shaped encoder-decoder network model was constructed, which solved the accuracy problem of multi-target recognition and grasping in unstructured scenarios, and enabled home service robots to achieve efficient multi-target recognition and intelligent grasping in complex environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2025-10-10
- Publication Date
- 2026-07-17
AI Technical Summary
Existing robotic grasping methods struggle to accurately identify stacked, occluded, and cluttered objects in unstructured scenarios, leading to inaccurate multi-target recognition and classification.
A boundary vector fitting-based approach is adopted. By establishing a multi-target grasping detection dataset, a U-shaped encoder-decoder network model with multi-scale asymmetric skip connections is constructed to output low-scale semantic feature mapping. The boundary vector fitting algorithm is used to predict the rotated bounding box and identify grasping classification information, and grasping parameters are calculated.
It improves the accuracy of multi-target recognition and classification in unstructured scenarios, enhances the robot's intelligent grasping ability in complex environments, and is suitable for lightweight deployment of home service robots.
Smart Images
Figure CN121482438B_ABST