一种机器人视觉作业的图像识别与理解方法
By establishing an environmental benchmark in the robot vision system, calculating the global anchoring vector and semantic momentum features, and combining differential hedging and local manifold measures, the semantic perception bias problem of the robot vision system in unstructured environments is solved, achieving stable and accurate recognition under extreme conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 伽利略(天津)技术有限公司
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-17
AI Technical Summary
In unstructured dynamic environments, the drastic switching of ambient light fields and local shadow interference in existing robot vision systems lead to semantic perception deviations in image recognition systems. Traditional methods, such as increasing the depth of convolutional layers or introducing attention mechanisms, increase the computational load and cannot effectively solve the manifold entanglement problem between environmental bias and target features, resulting in target loss or misidentification.
By acquiring the original image sequence, establishing an environmental baseline, calculating the migration trajectory and semantic momentum features of the global anchor quantity, using differential hedging processing to filter out background noise in the high-dimensional feature space while preserving the semantics of the target boundary, and combining local manifold measure to adjust the hedging damping coefficient, nonlinear hedging processing is achieved.
It maintains the stability and accuracy of recognition under extreme conditions, eliminates semantic drift, improves the ability to recognize subtle edge differences between the target and the background, and ensures that the robot maintains the stability of semantic perception under high-speed operation or light field flickering conditions.
Smart Images
Figure CN122135327B_ABST