Single-person attitude estimation method based on multistage prediction feature enhanced convolutional neural network

Abstract

The invention provides a single-person attitude estimation method based on a multistage prediction feature enhanced convolutional neural network. The method comprises the following steps: obtaining atraining set and a test set; grading the human skeleton points; constructing a multi-stage prediction feature enhancement convolutional neural network; training the multi-stage prediction feature enhanced convolutional neural network; and obtaining a single-person attitude estimation result based on the trained multi-stage prediction feature enhanced convolutional neural network. According to themethod, a multi-stage prediction feature enhancement convolutional neural network is adopted, the difficulty degree of human body posture skeleton point prediction is distinguished, and a parameter-free feature enhancement module is adopted, so that the extracted skeleton point features are finer. Meanwhile, the parameter quantity of the model is greatly reduced, and the accuracy and speed of single-person posture estimation are effectively improved.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and relates to a single-person attitude estimation method, in particular to a single-person attitude estimation method based on a multi-level prediction feature enhanced neural network, which can be used in various fields of human behavior analysis technology. Background technique [0002] With the rapid development of modern information technology, the use of human behavior analysis technology to quickly and accurately analyze human behavior in big data has a very wide range of applications, such as security monitoring, motion analysis and human-computer interaction. Accurate acquisition of human body posture skeleton points is the basis of human behavior analysis technology. Compared with using Kinect depth camera to obtain human body posture information, human body posture estimation algorithm can directly obtain human body skeleton points from images obtained by RGB cameras, which is e...

AI Technical Summary

Problems solved by technology

The implementation scheme is to correct the result of pose estimation at this stage by continuously using multiple cascaded stacked convolutional layers to fuse the characteristics of the input image to the output of each stage. In the process of model training, the skeleton points of human pose are not distinguished. The difficulty of detection, easy skeleton points occupy too many learning opportunities, which is not conducive to the learning of difficult skeleton points, making the feature representation of human pose skeleton points inaccurate, resulting in low accuracy of single-person pose estimation

In addition, the existing single-person pose estimation algorithm will obtain more accurate human pose results through at least six stages of feature fusion, and the number of model parameters will increase dramatically, resulting in a slowdown in the speed of single-person pose estimation.

Images