Human body three-dimensional joint point prediction method based on grouping regression model

Abstract

The invention discloses a human body three-dimensional articulation point prediction method based on a grouping regression model. The method comprises the following steps: collecting human body 2d articulation point detection data; inputting the 2d joint point coordinates into a regression network with the same structure, obtaining 3d joint positions in different groups, and merging the obtained key three-dimensional positions into an integral joint vector; constructing a joint point self-constraint network and a joint group self-constraint network through BiLSTM, and accumulating 3d joint points output by the two self-constraint networks to obtain a fine-tuned 3d prediction joint; and calculating the Euclidean distance between the 3d predicted joint and the 3d joint through a loss function. According to the invention, a grouping regression structure is adopted in combination with the characteristic of human limb joint movement independence, four limbs and a trunk are divided into different joint groups, the 3d positions of joint points in each group are predicted respectively, and meanwhile, in order to make a prediction result closer to a real human body posture, a human body joint self-constraint network is designed by utilizing BiLSTM to adjust the prediction result, so that the accuracy is improved.

Description

technical field [0001] The invention belongs to human body posture estimation technology, and in particular relates to a method for predicting three-dimensional joint points of a human body based on a grouping regression model. Background technique [0002] 3D human pose estimation is a challenging topic in the field of computer vision, which can be applied in fields such as virtual reality, action recognition, and human-computer interaction. The difficulty lies in that the collected image is a two-dimensional signal, and it is difficult to extract the depth information of the human body. Early methods used contour, shape, SIFT, HOG and other invariant features to reconstruct human poses. Although it is easy to implement, it has defects such as high algorithm complexity and low efficiency. In recent years, the theory of deep learning has been applied to 3D human pose estimation, and good results have been achieved. There are mainly two methods: [0003] 1. Learning an end...

AI Technical Summary

Problems solved by technology

[0007] To sum up, the existing technical solutions cannot conform to the actual movement characteristics of human limbs, and the obtained process needs to calculate and store a large amount of data, which is costly and time-consuming, and the final effect is not real and reliable.

Images