3D human body posture estimation model building method based on single-frame image and application of 3D human body posture estimation model building method

Abstract

The invention discloses a 3D human body posture estimation model building method based on a single-frame image and application thereof, and belongs to the field of 3D human body posture estimation.The 3D human body posture estimation model building method comprises the steps that a trained 2D human body posture estimation network is obtained, and 2D human body postures corresponding to RGB images in a training data set are estimated through the trained 2D human body posture estimation network; a back projection network is established, and the joint point grouping module is used for dividing n joint points into m joint point groups according to motion constraints; the m estimation branches are respectively used for estimating 3D coordinates of each joint point in the m joint point groups; the articulation point splicing module is used for splicing articulation point 3D coordinates corresponding to the m articulation point groups to obtain an estimation result of a 3D human body posture; the 2D human body posture estimation result corresponding to each RGB image is taken as input, the corresponding 3D human body posture actual value is taken as a label, and a back projection network is trained; and the two trained networks are connected in series to obtain a 3D human body posture estimation model based on the single-frame image. According to the invention, the estimation precision of the 3D human body posture can be improved.

Description

technical field [0001] The invention belongs to the field of 3D human body posture estimation, and more specifically relates to a method for establishing a 3D human body posture estimation model based on a single frame image and its application. Background technique [0002] The 3D human pose estimation task refers to the prediction of the three-dimensional coordinates of human joint points through RGB images or videos. This task has important application value in scenarios such as behavior recognition, human-computer interaction, and intelligent monitoring. [0003] The current 3D human pose estimation methods can be classified from the input form into 3D human pose estimation based on multi-view images, 3D human pose estimation based on video and 3D human pose estimation based on single frame images. The method based on multi-view image or video has more advantages in terms of model accuracy and generalization ability than the model that takes a single frame image as input...

AI Technical Summary

Problems solved by technology

However, since the 2D pose to 3D pose itself is a pathological process of solving the back projection, the same 2D pose back-projected back to the 3D space may correspond to multiple 3D poses, so it is still very challenging to predict the 3D pose using the 2D pose of a single frame image.

In the training process of the 3D pose estimation model, most of the current methods usually regard all the relevant nodes of the human body as a whole to learn a unified mapping from 2D coordinates to 3D coordinates, and do not fully consider the differences in the degrees of freedom of joint points during human motion, which affects the Accuracy of 2D pose reconstruction to 3D pose

[0006] Overall, the estimation accuracy of the existing two-stage mode 3D human pose estimation method needs to be further improved

Images