3D attitude estimation method based on graph cavity convolution encoder and decoder

Abstract

The invention relates to the field of computer vision, in particular to a 3D attitude estimation method based on a graph cavity convolution encoder and decoder. The method comprises the following steps: S1, selecting a training data set; S2, constructing a graph cavity convolution encoder and decoder model; S3, preprocessing the training data set; S4, carrying out initialization operation on the graph cavity convolution encoder and decoder model; S5, training a graph cavity convolution encoder and decoder model; S6, verifying a graph cavity convolution encoder and decoder model on the selected training data set. According to the method, the multi-scale context information can be effectively extracted, the overall and global range connection can be accurately captured, the information is very helpful for 3D attitude estimation, and the prediction performance of 3D attitude estimation can be greatly improved.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a 3D attitude estimation method based on a graph hole convolution encoder and decoder. Background technique [0002] For decades, Human Pose Estimation has attracted much attention in the computer vision community. It is a key step in understanding the behavior of characters in images and videos. Human pose estimation includes 2D pose estimation and 3D pose estimation. Among them, 2D pose estimation is mainly 2D human body joint point coordinates are estimated from images, and 3D pose estimation aims to regress from 2D joint point coordinates (or 2D images) to 3D pose estimation. 3D pose estimation is now attracting more and more attention in many computer vision fields , such as intelligent monitoring, human-computer interaction, video understanding, and VR, etc. In this task, the data used is skeleton data, a series of 2D coordinates of human joint points. Compared with RGB data,...

AI Technical Summary

Problems solved by technology

[0006] Based on the above problems, the present invention provides a 3D pose estimation method based on a graph-caused convolutional encoder-decoder, which solves the problem of ignoring multi-scale context information and semantic information in existing methods based on graph neural networks and extracting non-local Information ignores the problem of positional encoding information with rich semantic information

Images