Unsupervised pedestrian re-identification method based on hierarchical clustering and difficult sample triple

Abstract

The invention relates to an unsupervised pedestrian re-identification method based on hierarchical clustering and a difficult sample triple. The method comprises the following steps: 1) dividing a pedestrian image set into a training set and a test set; 2) initializing a label; 3) inputting the training set into a CNN model, extracting pedestrian features, and performing model training; 4) carrying out hierarchical clustering on the pedestrian features, and distributing the pedestrian features into clusters; 5) performing sampling according to the cluster category to obtain a new training set; 6) according to a difficult sample triple loss function, performing fine adjustment on the new training set; 7) according to the average contour coefficient of the training set, performing judging to return to the step 3) or judge the optimal CNN model; and 9) loading the test set into the optimal CNN model to obtain a pedestrian re-identification result. Compared with the prior art, the method not only can solve the problem that the number of individuals cannot be determined in pedestrian re-identification of a no-label data set, but also reduces the probability that different types of highly similar samples are clustered into one type, so that the identification performance of the model is greatly improved.

Description

technical field [0001] The invention relates to the technical field of unsupervised pedestrian re-identification, in particular to an unsupervised pedestrian re-identification method based on hierarchical clustering and triplets of difficult samples. Background technique [0002] With the continuous growth of the urban population, people pay more and more attention to social and public security issues. At present, many public places are covered with large-scale network cameras, which are an important guarantee for monitoring security. In order to improve the security intelligence level of network cameras, pedestrian re-identification technology is a research hotspot in the field of visual analysis, and has received extensive attention from the academic community. The purpose of pedestrian re-identification is to match pedestrians in a non-overlapping multi-camera network, that is, to confirm whether the pedestrian targets captured by cameras in different positions at differe...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide an unsupervised pedestrian re-identification method based on hierarchical clustering and difficult sample triplets in order to overcome the shortcomings of the unsupervised pedestrian re-identification method in the above-mentioned prior art that it is difficult to extract discriminative features from pedestrian images and the recognition accuracy is low. Supervised person re-identification methods

Images