A three-dimensional perception and positioning system for plate defects based on fusion of longitudinal and transverse laser grids and deep learning
By constructing a 3D perception system for board defects that integrates orthogonal laser meshes and deep learning, the problems of insufficient perception of board posture changes, laser image restoration artifacts, and encoder errors in existing technologies have been solved. This system achieves high-precision defect identification and 3D positioning, supporting accurate decision-making for robot grinding.
Patent Information
- Application Number
- CN202610549429.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-17
AI Technical Summary
Existing sheet metal defect detection technologies struggle to simultaneously meet the requirements of semantic accuracy in defect identification, robustness in 3D positioning, and cost-effectiveness in large-format deployment under dynamic transport conditions. In particular, parallel laser line arrays cannot detect changes in sheet metal posture, laser image restoration introduces artifacts, reliance on posture sensors increases system complexity, and encoder cumulative errors cannot be effectively compensated.
A 3D perception and localization system for sheet metal defects, based on the fusion of cross-laminar laser meshes and deep learning, is employed. This system constructs an orthogonal laser mesh to form a geometrically known, following physical reference frame. Combining deep learning semantic recognition and laser geometric measurement, it achieves automatic defect identification and precise quantification. The system includes a multi-source perception module, a laser stripe extraction module, a defect semantic recognition module, and a spatiotemporal fusion module. It utilizes the optical calculation of six-degree-of-freedom pose at the intersection points of the laser mesh, eliminating the need for attitude sensors. An algorithmic partitioning strategy is employed to shield against laser region interference, and encoder errors are verified and compensated in real time.
It achieves high-precision defect identification and 3D positioning in a high-speed dynamic conveying environment, avoids interference from laser stripes on semantic recognition, simplifies system calibration, reduces deployment costs, and provides accurate decision support for robot grinding.
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Citation Information
Patent Citations
A high-resolution, high-precision dual photoelectric encoder compensation subdivision device and method
CN107421569B
Industrial detection multi-line laser light stripe center line extraction method
CN117237433A
Compensation method for attitude change of linear laser three-dimensional reconstruction system
CN120823305A