Pedestrian re-identification method based on multi-scale pedestrian contour segmentation fusion

Abstract

The invention discloses a pedestrian re-identification method based on multi-scale pedestrian contour segmentation fusion. The method comprises the steps of firstly performing data preprocessing; secondly, extracting global features of the images and contour features of pedestrians, and fusing the two features; training the pedestrian re-identification network by adopting a label smooth loss function, so that network parameters are optimal; and finally, for a query set and a candidate set contained in the pedestrian re-identification data set, calculating an Euclidean distance between a specified object in the query set and each object in the candidate set, and then performing ascending sorting on the calculated distances to obtain a pedestrian re-identification sorting result. According to the method, features on pedestrian clothes are removed, the human body contour of the pedestrian is learned to identify the pedestrian, and pedestrian re-identification is carried out in combinationwith global features. No matter whether the pedestrian clothes are replaced, the pedestrian clothes can be well re-identified.

Description

Technical field [0001] The invention relates to the technical field of computer vision, in particular to a pedestrian re-identification method based on multi-scale pedestrian contour segmentation and fusion. Background technique [0002] Pedestrian re-identification is also called pedestrian re-identification (Re-ID), which is actually a technology that uses computer vision technology to determine whether there is a specific pedestrian in an image or video sequence. Specifically, it refers to identifying the identity of the pedestrian from the pedestrian images captured by different cameras . Given an image containing a target pedestrian (query), the ReID system tries to search for images that include the same pedestrian from a large number of pedestrian images (galleries), which is widely regarded as a sub-problem of image retrieval; given a pedestrian image under surveillance, search The image of the pedestrian under the cross-device. Designed to make up for the visual limita...

AI Technical Summary

Problems solved by technology

[0003] ReID has developed rapidly in the past two years, but compared with face technology, there are very few landing applications

In fact, it is not that the ReID model is not good enough, or that the accuracy of the data set is not high enough, but that compared with the face task, the ReID scene is more complicated, and there are some essential problems that have not been solved.

ReID remains a very challenging task due to a large number of uncontrolled sources of variation, such as significant changes in pose and viewpoint, complex changes in lighting, and poor image quality

[0004] The simplest and most urgent problem of occlusion, invisible light, and replacement of pedestrian clothing can make almost all existing ReID models extremely poor, which can be said to be invalid.

Images