Pedestrian re-identification method based on attention mechanism

Abstract

The invention provides a pedestrian re-identification method based on an attention mechanism, and the method is characterized in that the method comprises the following steps: performing data set preparation, data cleaning, preprocessing, and data enhancement; designing a feature extractor; designing a feature optimizer; designing a feature fusion device; designing a feature classifier; designing an attention mechanism on a channel domain; designing an attention mechanism in a spatial domain; enabling the synthesis network to train the input data; and adding feature perception between modals on the basis of context perception, and obtaining correlation weight parameter values between different features through learning, so that the effective distinguishing capability of the features is improved. According to the method, the detail information of the low-quality image is well reserved, the characterization capability of the features is improved, and the discrimination capability of the network pedestrian recognition task is enhanced.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a pedestrian re-identification method based on an attention mechanism. Background technique [0002] With the rapid development of modern society, social security measures and people's safety awareness are increasing day by day. The application of monitoring technology has also penetrated into all aspects of life. Whether it is in a crowded place or a place that requires security, general monitoring is all-round coverage. In the field of surveillance, computers obtain huge amounts of visual information through surveillance cameras, and at the same time use related algorithms to complete recognition tasks. The target detection algorithm is used to locate and identify pedestrians, motor vehicles, non-motor vehicles and other targets in the surveillance video. The monitoring tasks are completed through this algorithm, including monitoring target tracking, behavior recognit...

AI Technical Summary

Problems solved by technology

At present, the pedestrian re-identification method based on the convolutional neural network generally chooses to use the twin model, but the problem with the twin network is that it may only learn the label information of the image pair or triple image during the network training process, and cannot make full use of the annotation of the pedestrian target. information, the learned feature representation may be biased

Images