Posture estimation and human body analysis system based on multi-task deep learning

Abstract

The invention discloses a posture estimation and human body analysis system based on multi-task deep learning. The system comprises a human body detection subnet and a posture estimation and human body analysis combined learning subnet. An input image firstly passes through a human body detection subnet to obtain information such as a human body position and a mask, and an interference-free single-person image is extracted from a multi-person image according to the information; the method further includes performing attitude estimation and human body analysis joint learning on the interference-free single-person image to obtain an attitude estimation result and a multi-granularity human body analysis result; and finally, combining the single-person posture estimation result and the multi-granularity human body analysis result to the original image. Different human body instances are distinguished based on human body postures, and a better human body detection effect is achieved on multi-person images; according to the invention, the accuracy of two tasks of posture estimation and human body analysis can be improved; and a cascade network structure is adopted for a human body analysis task, so that the human body analysis accuracy can be effectively improved, and finer analysis granularity expansion is facilitated.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a posture estimation and human body analysis system based on multi-task deep learning. Background technique [0002] Pose estimation and human body analysis are hot research tasks in the field of computer vision: the goal of the pose estimation task is to predict the position of the key points of the human body in the image, and obtain the pose structure of each person; the goal of the human body analysis task is to segment the semantic parts of the body in the image , to obtain the human body part corresponding to each pixel. At present, most of the mainstream methods only study one of the tasks, do not make full use of the correlation between the two tasks, and do not consider the mutual occlusion of the human body in the actual scene: do the pose estimation task on the basis of the general target detection system, or in the The general object detection sys...

AI Technical Summary

Problems solved by technology

[0003] Most of the current mainstream methods in the present invention only study one task of pose estimation and human body analysis, and do not make full use of the correlation between these two tasks, and do not consider the mutual occlusion of human bodies in actual scenes. A multi-task joint learning system for human analysis, and a solution to existing difficulties

Images