A six-degree-of-freedom pose estimation method based on projection point coordinate regression

Abstract

The invention discloses a six-degree-of-freedom pose estimation algorithm based on projection point coordinate regression, which expands on the pose estimation problem based on the single-step target detection algorithm, and learns from the method of BB8 algorithm to return the projection point position, for each The detected object of interest is returned to the projected coordinates of the vertices of the 3D bounding box, and then the EPnP algorithm is used to calculate the six-degree-of-freedom pose parameters. The convolutional neural network proposed in the present invention can perform end-to-end training and prediction for the target of pose estimation, which improves the calculation efficiency of the algorithm and the accuracy of pose estimation, and can achieve the current best result without post-processing of pose correction. Excellent pose estimation accuracy can realize real-time processing, and in multi-target scenarios, only one EPnP calculation is required for each detected object instance, and there is no need to repeatedly run the algorithm multiple times, saving the running time of the algorithm.

Description

technical field [0001] The present invention relates to the technical field of digital image processing, and more specifically relates to the digital image processing technology of three-dimensional stereoscopic vision. Background technique [0002] Determining the three-dimensional translation and rotation transformation parameters (a total of six degrees of freedom) of an object in an image relative to a camera is a classic problem in the field of computer vision. In recent years, the development of emerging applications such as augmented reality, autonomous driving, and visual robots has required higher accuracy and speed for six-degree-of-freedom pose estimation, and academia has also conducted a lot of research work on this demand. In terms of the form of input data, existing algorithms can be mainly divided into pose estimation algorithms based on RGB images and pose estimation algorithms based on RGBD data. Among them, the pose estimation algorithm based on RGBD data...

AI Technical Summary

Problems solved by technology

[0006] The BB8 algorithm uses a multi-step processing method, resulting in low computational efficiency; multiple convolutional neural networks are trained separately, and cannot be comprehensively tuned for the pose estimation target, resulting in low pose estimation accuracy; the pose correction process involves image Rendering and other operations have high computational complexity, which leads to a slow running speed of the algorithm; for scenes with multiple objects, the BB8 algorithm needs to perform separate calculations for each object instance, resulting in a significant increase in the running time of the algorithm

Images