A pedestrian re-identification method, device, electronic equipment and readable storage medium

Abstract

The invention discloses a pedestrian re-identification method, device, electronic equipment and readable storage medium, belonging to the field of image processing. The method includes: the updated network model obtained by initializing the network parameters of the original network model includes a feature distribution alignment module, an aggregation Class module and noise label correction module; use source domain samples and target domain samples to train the feature distribution alignment module to obtain the alignment network model; train the clustering module based on the alignment network model and target domain samples to obtain the clustering results and each cluster The noise label corresponding to the result; based on the updated network model, target domain samples and each noise label, the noise label correction module is trained to obtain the target network model; the target network model is used to perform pedestrian re-identification on the query image. The accuracy of noise labels is improved through feature level alignment and iterative clustering, and the accuracy of noise labels is improved through the noise correction process, thereby improving the recognition accuracy of the target network model.

Description

technical field [0001] The invention belongs to the field of image processing, and more specifically relates to a pedestrian re-identification method, device, electronic equipment and readable storage medium. Background technique [0002] Pedestrian re-identification is a basic task in automatic video surveillance, and its purpose is to complete the matching of the appearance results of the same pedestrian under different cameras. By acquiring an image of a pedestrian in one camera, the model generates corresponding feature vectors for comparison with feature vectors from another camera view or images observed at different time points. The goal is to determine a person's past or present location by matching images. [0003] The actual pedestrian re-identification research work mainly consists of two parts: feature extraction (representation of pedestrian appearance features) and distance measurement (similarity comparison between pedestrians). In the process of learning pe...

AI Technical Summary

Problems solved by technology

Even if some methods take into account the target domain information, the mining of this information is not deep enough, or a large amount of noise is introduced, so that a good recognition effect cannot be achieved.

Images