Human body posture estimation method based on relation modeling

Abstract

The invention discloses a human body posture estimation method based on relation modeling, and the method comprises the following steps: 1, giving a section of video containing N frames, i.e., RPSTN adopts continuous T frames as the input, and giving an initial posture generated by P. Compared with the existing method, the RPSTN provided by the invention has certain robustness for the occlusion problem, and the RPSTN has a certain robustness for the occlusion problem; on one hand, the JRE module can infer the positions of shielded joint points in space through learned attitude structure information, namely the relation between the joint points, on the other hand, the JRPSP module can transmit attitude information in an unshielded frame to a shielded frame to help to position the attitude, and in the process of generating a joint point heat map, the problems that in an existing method, single joint point-by-point positioning is adopted, the relation between the joint points is ignored, a human body is an organic whole, all the joint points are correlated in the movement process, and structural information of the human body can be damaged in a point-by-point positioning mode in the existing method are solved.

Description

technical field [0001] The invention belongs to the technical field of human body posture estimation, and in particular relates to a human body posture estimation method based on relational modeling. Background technique [0002] A relation is actually the state or content of a relational schema at a certain moment. That is, the relational schema is the type and the relation is its value. The relational schema is static and stable, while the relation is dynamic and changes over time, because relational operations are constantly updating the data in the database. But in practice, relational schemas and relational systems are often referred to as relations, and readers can distinguish them from the context. [0003] The attitude estimation problem is to determine the orientation of a three-dimensional target object. Pose estimation has applications in many fields such as robot vision, motion tracking, and single-camera calibration. Different sensors are used for pose estim...

AI Technical Summary

Problems solved by technology

Although the current human pose estimation based on deep learning has made great progress, most of the methods have a common problem, that is, in the process of generating joint point heat maps, the existing methods are point-by-point positioning of individual joint points, while ignoring the The connection between joint points, the human body is an organic whole, and each joint point is related to each other during the movement process, and the existing method of point-by-point positioning will destroy the structural information of the human body

Images