Multi-person posture estimation method based on adversarial learning

Abstract

The invention discloses a multi-person posture estimation method based on adversarial learning. The multi-person posture estimation method comprises the following steps: employing a public data set with a multi-person key point coordinate label as a training set, and carrying out the edge information enhancement preprocessing of a training set image; preprocessing the key point coordinate tags inthe training set, and making a corresponding key point hotspot map and an overall skeleton hotspot map; constructing a double-branch key point feature extraction sub-network; constructing an A-HPose network generator part; constructing an A-HPose network discriminator part; performing relay supervision loop training on the A-HPose network by using the training set to obtain network model parameters; and carrying out post-processing on the network output hotspot map, carrying out search classification on key points in the key point hotspot map according to the skeleton hotspot map to obtain a key point position of each person in the plurality of persons, and estimating the postures of the plurality of persons. The multi-person posture estimation method has the beneficial effect of quickly and accurately detecting human body key point features.

Description

technical field [0001] The invention relates to the technical fields of computer vision and image recognition. More specifically, the present invention relates to a multi-person pose estimation method based on adversarial learning. Background technique [0002] In recent years, with the rapid development of mobile Internet technology, image and video data containing visual information have shown explosive growth. How to find visual target objects that may contain semantic content from massive images and videos is of great significance. Human Pose Estimation (HumanPose Estimation) is the process of detecting the key parts or main joints of the human body in a given image or video, and finally outputting all or partial limb-related parameters of the human body (the relative positional relationship of each joint point), such as the human body Outline, position and orientation of the head, position and part category of human joints, etc. Human pose estimation research involves...

AI Technical Summary

Problems solved by technology

CNN can extract deeper and richer data hidden information through multi-layer iterative convolution, but the occlusion of limbs in the image and the presence of clothes will also affect the visibility of human posture

Images