Multi-person posture estimation method based on global information integration

Abstract

The invention discloses a multi-person posture estimation method based on global information integration. The multi-person posture estimation method comprises the following steps: carrying out pre-processing on an input image; generating a group of human body boundary frames through a human body detector, and inputting the obtained human body boundary frames into a'feature coding + posture decoding 'module to carry out model training; sequentially predicting the positioning of the key points of each person, and generating a plurality of key point heat maps to represent the position confidenceof each key point; and finally, eliminating redundant attitude estimation through an attitude non-maximum suppression module to obtain a final human body attitude.. By combining different normalization strategies with multi-layer information fusion, the accuracy of multi-person posture estimation can be remarkably improved, false connection can be effectively reduced by adopting a hyperedge geometric constraint strategy, and a posture estimation method which is difficult to encounter in scale change, shielding and complex multi-person scenes can be effectively improved.

Description

technical field [0001] The invention relates to a multi-person pose method, in particular to a multi-person pose estimation method based on global information integration, and belongs to the technical field of image processing. Background technique [0002] Trying to make computers have the ability to automatically understand the human behavior information contained in images or video sequences has always been a hot topic in many research fields related to machine learning. Human body pose estimation is an important basis for these tasks, and has a wide range of applications in the fields of behavior recognition, human-computer interaction, human re-identification, audio-visual entertainment, etc. Human pose estimation refers to the process of locating body key points (head, shoulders, elbows, wrists, knees, ankles, etc.) from images, and determining the position and direction of different human body parts in the image through image analysis. , is the basis of human action ...

AI Technical Summary

Problems solved by technology

The two methods have their own advantages and disadvantages. The former requires a matching algorithm. Although it can achieve real-time, the accuracy is not high; the accuracy of the latter pose estimation depends to a large extent on the quality of the detected bounding box, and the more people, The calculation cost is greater, but the accuracy is higher than that based on the partial framework

[0005] Due to the complexity and changeability of natural images, human pose estimation faces many challenges: for images with complex backgrounds or weak lighting conditions, the appearance similarity between the human body and the background may be high; the appearance features of the same part of different human bodies are often different. The big difference is mainly due to the different lighting conditions of different pictures, different clothing and body shapes of different human bodies, and different sports models; the appearance of human body parts may not be complete, mainly due to mutual occlusion between human body parts or being occluded by other objects; The part search space is too large, because the human body part may be located in any area and angle of the picture without any prior; the complexity increases with the increase of real-time people

Images