Human body posture estimation method based on motion feature constraints

Abstract

The invention relates to a human body posture estimation method based on motion feature constraints. According to the method, a video is divided into a plurality of human body space-time windows, image features of each frame under the windows are extracted, human body motion features of the windows are extracted according to the image features contained in the windows, and then human body postureestimation of a single frame of picture is restrained through the motion features. According to the invention, the architecture of the full convolutional neural network is adopted, and the dynamic convolution is used to adaptively adjust the single-frame attitude estimation according to the situation information contained in the video, so that the common problems of motion blur, limb occlusion andthe like in a human body attitude estimation task can be better solved, the attitude estimation can be better carried out on the video, and the human body posture estimation accuracy in a video sceneis improved.

Description

technical field [0001] The invention belongs to the field of human body pose estimation, and in particular relates to a human body pose estimation method based on motion feature constraints. Background technique [0002] Human pose estimation is an interesting research field in computer vision. It has important application value in security monitoring, automatic driving, human-computer interaction, video understanding and other fields. The goal of human pose estimation is to estimate the pose of the human body in a picture or video image, by locating the key parts of the human body, and then connecting these key points to realize the prediction of the human body pose. [0003] Current approaches to human pose estimation generally fall into two categories: top-down and bottom-up. From top to bottom, starting from the whole world, first use the target detection technology to locate the position of each person in the picture or video image, and then perform single-person pose ...

AI Technical Summary

Problems solved by technology

The bottom-up method focuses on human body parts, first extracts the location information of key parts of all people in the picture and the affinity between parts, and then groups and clusters key points according to the topology of human body poses, and finally achieves The attitude estimation of all people, the bottom-up running time is not affected by the number of people and can run faster, but the accuracy is not high

[0004] Most of the current human pose estimation methods are for pictures. However, video is the main carrier in real application scenarios. Existing methods decompose video into single-frame pictures, and then perform human pose estimation frame by frame. This approach often ignores single frames. The difference between images and static pictures, for example, there are often situations such as motion blur and lens shake in a single frame, and the rich correlation information between video frames is not fully considered, resulting in the inaccuracy of human pose estimation in video Satisfactory

Images