Unsupervised pedestrian re-identification method based on online hierarchical clustering

Abstract

The invention discloses an unsupervised pedestrian re-identification method based on online hierarchical clustering. Comprising the steps of collecting pedestrian sample pictures and forming a training set; inputting the training set into a pedestrian re-identification model and outputting features of all pedestrian sample pictures, performing online hierarchical clustering on the features of all the pedestrian sample pictures by using an online hierarchical clustering method, then generating a pseudo tag corresponding to the training set, performing training and iterative training on the pedestrian re-identification model according to the pseudo tag, and obtaining a trained pedestrian re-identification model; selecting a to-be-detected pedestrian picture and extracting features of the to-be-detected pedestrian picture; and calculating the cosine similarity between the features of the to-be-detected pedestrian images and the features of the remaining to-be-detected pedestrian images in the to-be-detected pedestrian data set, and according to a preset similarity threshold, retaining all to-be-detected pedestrian images higher than the preset similarity threshold to realize pedestrian re-identification. According to the invention, an unsupervised pedestrian re-identification task can be completed, and the accuracy is high.

Description

technical field [0001] The invention relates to an unsupervised pedestrian re-identification method, in particular to an unsupervised pedestrian re-identification method based on online hierarchical clustering. Background technique [0002] Pedestrian re-identification methods can search for specific pedestrians in large data sets, and have a wide range of applications in real life. With the development of the times, the number of pictures in the dataset has grown exponentially. Due to the high cost of manual labeling, and the speed of simultaneous labeling is far behind the growth rate of data volume, unsupervised person re-identification methods are the future development trend. The unsupervised method can make full use of the advantages brought by the large amount of data, but the current bottleneck is that the accuracy of generating pseudo-labels is not high enough. At the same time, the current method has the problem that the pseudo-label update does not match the feat...

AI Technical Summary

Problems solved by technology

The unsupervised method can make full use of the advantages brought by the large amount of data, but the current bottleneck is that the accuracy of generating pseudo-labels is not high enough. At the same time, the current method has the problem that the pseudo-label update does not match the feature update rate, so training It is difficult to obtain a high-accuracy model after

Images