Pedestrian attribute identification method based on graph convolution

Abstract

The invention discloses a pedestrian attribute identification method based on graph convolution. The method comprises the following steps: firstly, extracting a human body key point heat map through HRNet, generating a global feature map through a residual network, and combining the global feature map with the human body key point heat map to generate human body key point local features; then grouping the key points of the human body, and learning the joint features of the key points of the human body by using a self-adaptive graph convolution module; and finally, classifying and recognizing semantic features of pedestrians through a pedestrian attribute recognition module. According to the method, the self-adaptive graph convolution module is utilized to make full use of the relation between the inherent structures of the human body, the problem that a traditional pedestrian attribute recognition method is poor in prediction effect is effectively solved, and compared with an existing pedestrian attribute recognition method, efficiency and prediction accuracy can be greatly improved.

Description

technical field [0001] The invention relates to a pedestrian attribute recognition method based on graph convolution, which belongs to the technical field of artificial intelligence and computer vision. Background technique [0002] The video surveillance market is experiencing explosive growth. In the past few years, the deployment of high-definition network cameras in various surveillance application scenarios has become more and more common, and video analytics functions are also increasingly embedded. With the development of computer vision technology, the technology of target (pedestrian) detection in surveillance video has become very mature, and people are eager to extract more useful information about detected pedestrians, such as gender characteristics, age characteristics, appearance characteristics, etc. These pedestrian features are collectively referred to as pedestrian attributes. Pedestrian attributes are human-searchable semantic descriptions, which have app...

AI Technical Summary

Problems solved by technology

[0003] However, due to factors such as viewing angle changes, illumination changes, and occlusions, pedestrian attribute recognition is still challenging.

Images