Six-degree-of-freedom object attitude estimation method and system based on symmetric perception

Abstract

The invention provides a six-degree-of-freedom object attitude estimation method and system based on symmetric perception, and is suitable for symmetric objects, and the method comprises the steps: segmenting a to-be-estimated object in RGBD data through employing a preset instance segmentation algorithm; inputting into a preset neural network, and predicting the attitude of the object; using an object symmetric perception algorithm as a loss function to carry out loss calculation, and reversely training the neural network, the object symmetric perception algorithm being a symmetric invariant attitude distance measure with symmetric perception, finding expressions with the same attitude characterization by using a rotational symmetry mode, and calculating the attitude distance of the neural network; and calculating the loss cost based on the symmetry characteristic of the to-be-estimated object. According to the scheme of the invention, for the rotation ambiguity problem of the symmetric object, the proposed symmetric perception expression algorithm can accurately estimate the attitude of the symmetric object, effectively eliminate the local optimum condition of the symmetric object, and guide the neural network to enter a correct state, so that the network learns correct parameters.

Description

technical field [0001] The present invention relates to the field of 6D attitude estimation, in particular to a method and system for estimating the attitude of a six-degree-of-freedom object based on symmetry perception. Background technique [0002] The purpose of 6D pose estimation is to detect objects in the scene and estimate the rotation and translation of the object relative to the canonical frame. Efficient and accurate 6D pose estimation is a key technology for real-time interactive applications, such as augmented reality, autonomous driving, and robot applications. [0003] Designing a robust pose estimation algorithm is challenging due to changes in illumination, data noise, and incomplete occlusions in real scenes. In recent years, thanks to the powerful feature extraction and fitting capabilities of deep learning, data-driven pose estimation algorithms have shown excellent performance. The existing attitude estimation algorithms based on deep learning can be d...

AI Technical Summary

Problems solved by technology

[0005] However, none of the existing solutions can deal with the rotation ambiguity of symmetrical objects, so the algorithm can only be applied to asymmetric objects, which greatly limits the application scenarios of the algorithm.

Even if the unambiguous ADD(S) is used as the cost loss function, ADD(S) is not accurate in describing the rotation error, which will cause the network to fall into a local optimal state, and it is difficult to guide the network into the correct state

Images