Semi-supervised object pose estimation method combining generated data and unlabeled data

Abstract

The invention relates to a semi-supervised object pose estimation method combining generated data and unlabeled data. The method comprises the following steps: 1) generating point cloud data with a pose label, namely the generated data; 2) obtaining a color image and a depth image of a target object without labels, inputting the color image into the trained instance segmentation network to obtain an instance segmentation result, and obtaining a point cloud of the target object from the depth image according to the segmentation result, namely real data without labels; 3) in each training period, performing supervised training on the pose estimation network model by using the generated data, and performing self-supervised training on the pose estimation network model by using real data without labels; and 4) after each training period is finished, calculating the accuracy of the pose estimation network model by adopting part of real data. Compared with the prior art, the method mainly solves the problem that the 6D pose label is difficult to obtain, and accurate object pose estimation can be realized only by using synthetic data and unlabeled real data.

Description

technical field [0001] The invention relates to the field of robot vision, in particular to a semi-supervised object pose estimation method combining generated data and unlabeled data. Background technique [0002] The object pose estimation technology based on computer vision is the key technology to realize robot grasping and dexterous operation. It can improve the environment and task adaptability of the robot, broaden the application field of the robot, and improve the application of the robot in intelligent manufacturing, warehousing logistics, home service and other scenarios. The flexibility and application performance in it are of great significance. In addition, this technology also has broad application prospects in the fields of autonomous driving, augmented reality, and virtual reality. [0003] In recent years, with the vigorous development of deep learning technology, object pose estimation based on deep learning has achieved good results. In unstructured sce...

AI Technical Summary

Problems solved by technology

Although pose labeling is not required, this method still needs to use the generated color images for supervision, the influence of domain differences has not been eliminated, and the accuracy of this method cannot meet the needs of practical applications

Images