6D pose estimation method fusing point cloud local features

Abstract

The invention relates to a 6D pose estimation method fusing point cloud local features, and the method comprises the following steps: firstly, dividing an image obtained by a three-dimensional camera into an RGB image and a depth image, then extracting the feature information of the input RGB image, and analyzing the color information of the RGB image; secondly, for a depth point cloud image, densely connecting points in a local area into a local point network to know the influence of each point on other points, so as to adjust the features of the points; thirdly, carrying out pixel-by-pixel dense fusion on the obtained color information and point cloud depth information, and generating a pixel-by-pixel estimated pose by combining a fusion result with a global feature of dense fusion; and inputting the estimated poses of all the pixel points into a final multi-layer perceptron, and then carrying out average maximum pooling on the confidence coefficient, thereby predicting the 6D pose of the object. According to the method, the local feature information of the point cloud is effectively incorporated into the point elements, so that the capability of describing the local neighborhood by the elements is enhanced, and the performances of 6D pose estimation stability, accuracy and the like are remarkably improved.

Description

technical field [0001] The invention relates to the technical field of robot environment perception, in particular to a 6D pose estimation method for merging local features of point clouds. Background technique [0002] Due to its high flexibility, robot environment perception technology has penetrated into various fields, such as intelligent logistics, defect detection, etc. 6D pose estimation refers to the rotation and translation transformation relationship between the camera coordinate system and the target object coordinate system. There are 6 pose quantities including position and rotation angle. It is used in applications such as autonomous driving, industrial intelligent robots, and AR. play a vital role in the field. [0003] From the perspective of practical application, most of the 6D pose estimation methods currently used in industrial scenes can be divided into three types. The first type is based on matching methods, which are suitable for target objects with ...

AI Technical Summary

Problems solved by technology

[0006] In summary, due to the inherent defects of two-dimensional vision and traditional algorithms, the accuracy of pose estimation is not high, the real-time performance and robustness are poor, and it cannot cope with more complex industrial environments.

Images