Pedestrian attribute recognition method based on generative adversarial learning

Abstract

The invention relates to a pedestrian attribute recognition method based on generative adversarial learning, which belongs to the field of deep learning and image processing, and comprises the following steps: S1, establishing an encoder, a generative network, a discrimination network and an attribute recognition network; s2, preprocessing the real database image; s3, extracting a real image pairfeature to train the generation network, and performing authenticity judgment on the image generated by the generation network by adopting a judgment network; s4, training an attribute recognition network; and S5, performing training adjustment on the overall framework, and utilizing the trained attribute recognition network to extract pedestrian attributes in the image. The invention discloses anadversarial learning method, a pseudo sample image is generated by interacting attribute features between different pedestrians, the sample space of pedestrian attribute training is expanded, data distribution is balanced, the robustness of pedestrian attribute recognition is enhanced, and furthermore, an image generation network and an attribute recognition network are end-to-end networks, so that the attribute recognition network can be better trained.

Description

technical field [0001] The invention relates to a pedestrian attribute recognition method, which belongs to the field of deep learning and image processing, and is especially suitable for pedestrian attribute recognition based on generative confrontation learning. Background technique [0002] With the advancement of the Safe City and Skynet projects, the number of video cameras in the city is increasing exponentially. However, video, as a typical unstructured data, is difficult to use directly. For example, in the field of security, we often need to find target persons in video surveillance, or when analyzing important information such as gender, age structure, and clothing attributes of visitors in surveillance videos such as shopping malls and communities, it often needs to be processed by manual identification. . However, it is time-consuming and labor-intensive to manually process a large number of videos, especially when viewing for a long time, the efficiency and acc...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the present invention provides a pedestrian attribute recognition method based on generative adversarial learning, using the adversarial learning neural network to generate more pseudo-training samples, which can effectively expand the sample space and solve the problems of data imbalance and data labeling from the root

Images