Human body posture prediction method and system based on attention mechanism fused with multi-stream graph

Abstract

The invention relates to a human body posture prediction method and system based on an attention mechanism fused with a multi-stream graph neural network, and the method comprises the steps: S1, obtaining a three-dimensional position data sequence of a human body key joint for training, and dividing the three-dimensional position data sequence into an input sequence and an output sequence according to the lengths of a preset input sequence and an output sequence; constructing graph data according to the input sequence; S2, constructing a neural network model based on the attention mechanism fusion multi-stream graph; inputting graph data into the model for training to obtain a trained model; and S3, obtaining a three-dimensional position data sequence of human body key joints for prediction, constructing graph data, and inputting the trained attention mechanism-based fusion multi-flow graph neural network model to obtain a predicted value of a human body posture. According to the method provided by the invention, a plurality of graph models are constructed based on the human body joint position data and the structural features, modeling of a human body motion system is realized, human body postures are predicted, and relatively high accuracy is achieved.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a human body posture prediction method and system based on an attention mechanism fused with a multi-flow graph neural network. Background technique [0002] In recent years, with the widespread use of low-cost consumer-grade depth cameras, the low-cost and real-time acquisition of human three-dimensional motion poses has become possible. Therefore, human pose prediction has become a hot issue at the intersection of graphics and computer vision. It has broad application prospects and rich application scenarios in the fields of medical and autonomous driving. [0003] In the field of robotics, exoskeleton robot-related technologies are a hot topic of research, and have very important applications in aerospace. Completion of cabin tasks and emergency fault handling will play an extremely important role. The human body posture prediction algorithm recognizes an...

AI Technical Summary

Problems solved by technology

Early deep learning models were mainly based on cyclic neural networks and convolutional neural networks, but they had certain defects: cyclic neural networks emphasized the temporal relationship of sequences, ignoring spatial information; the work of convolutional neural networks allowed single-frame bone data to be constructed into a Dimensional vector, which treats the sequence as a two-dimensional matrix, focuses on the position change of a single joint over time, and ignores the correlation between the joints of the human body, and cannot make full use of the topological structure information of the human body itself

Therefore, constructing the adjacency matrix only according to the spatial adjacency relationship of human joints cannot make full use of the prior information on the structure and kinematics of human joints.

Images