Detection and Correction Method of Packaging Box Segmentation Errors Based on Vision and Active Interaction

Abstract

The invention belongs to the technical field of logistics, and discloses a visual and active interaction-based detection and correction method for a packing box segmentation error. Carrying out regionextraction based on a target detection model result; performing Suspected error segmentation detection based on a two-dimensional visual algorithm; performing Error segmentation judgment based on anactive interactive strategy; Correcting error segmentation and calculating a correct grabbing position; based on Two-dimensional vision-based algorithm, the possible problems of under-segmentation andover-segmentation of the prediction result of the target detection model are pre-judged; a gripper of the robot to perform trial gripping is guided on the target, the size and position change of thebox body are tracked, detecting whether the comparison result is matched with a prior segmentation result or not, if yes, continuing to complete gripping on the target, and if not, performing correction on error segmentation. The visual-based interactive strategy can effectively correct the packaging box which is wrongly segmented, the risk that the packaging box is broken down is avoided, and therobustness of the robot for accurately grabbing the packaging box is greatly improved.

Description

technical field [0001] The invention belongs to the technical field of logistics, and in particular relates to a detection and correction method for packaging box segmentation errors based on vision and active interaction. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: [0003] In the modern logistics industry, the unpacking and sorting of packing boxes is a heavy and repetitive work. Many robotics companies are trying to replace humans with robots to complete this part of the task. The primary problem of dismantling and sorting by robots instead of humans is: how to accurately segment and identify dense packing boxes, so as to guide the robot to accurately grasp. In practice, the size of the packaging box varies widely, the angle and posture of the placement are uncertain, and the gap between the packaging boxes is small. These factors lead to a high failure rate of the packaging box segmentation. The exi...

AI Technical Summary

Problems solved by technology

The primary problem of dismantling and sorting by robots instead of humans is: how to accurately segment and identify dense packing boxes, so as to guide the robot to accurately grasp

Although the current best target detection model has a very good effect on the problem of unpacking and sorting boxes, there are still two problems in practice: First, when the texture pattern on the box appears a dividing line that runs through the box, it is easy to cause The over-segmentation phenomenon predicted by the network is to identify one box as multiple boxes; the second is that the boxes are placed densely and the gap between the boxes is too small, which will make the boundaries between the boxes unclear, which will easily lead to the under-segmentation phenomenon predicted by the network. Multiple boxes are recognized as one box

[0007] Difficulty: The unpacking and sorting scene of logistics packaging boxes is relatively complex, and the items are too dense, often dozens of boxes are closely stacked, and the degree of discrimination is low; the texture patterns of boxes vary widely, and the texture and box edges are easy to confuse and difficult to identify

The above factors make it very difficult to accurately segment the boxes in the demolition scene, and the failure rate is high

Images