A Pedestrian Attribute Recognition Method Based on Multi-scale Spatial Correction

Abstract

The invention discloses a pedestrian attribute recognition method based on multi-scale spatial correction. The method includes acquiring pedestrian image data and performing preprocessing; constructing a pedestrian attribute recognition network including a feature pyramid structure, a space correction module and a multi-scale feature fusion module Model, using the preprocessed pedestrian image data for model training; using the trained pedestrian attribute recognition network model to identify pedestrian attributes in the pedestrian image to be recognized. The present invention uses the spatial correction module to simultaneously input the adjacent features in the feature pyramid, uses the small-scale high-level semantic information to guide the large-scale low-level feature conversion; and uses the feature fusion module to fuse the features of all scales in the feature pyramid to establish Multi-scale spatial correlation dependencies can significantly improve the recognition effect of small targets and low-resolution images.

Description

technical field [0001] The invention relates to the technical field of pedestrian attribute recognition, in particular to a pedestrian attribute recognition method based on multi-scale space correction. Background technique [0002] With the society's emphasis on the security field and the continuous development of the security field, the integration of security and AI has become increasingly close. Among them, intelligent video analysis has received extensive attention. Video images are the most extensive information carrier in today's society, especially in video surveillance, playing an important role in information collection and recording. Video is typical unstructured data, which requires intelligent algorithms to assist in structured analysis. In video surveillance, people are the main body of security work. Realizing the effective identification of pedestrian targets and their attribute characteristics in security work will greatly improve the response ability of se...

AI Technical Summary

Problems solved by technology

[0004] Local area features are very important for fine-grained attribute classification, but because fine-grained attributes (such as shoes, glasses, etc.) account for a small proportion in the image, recognition is more difficult

At the same time, high-quality surveillance cameras are expensive, and the resolution of images is usually very low. In addition, there are problems such as occlusion of portraits and lens blurring in open environment applications, making it more difficult to effectively extract features of small areas.

Images