Unsupervised pedestrian re-identification method for enhancing sample data

Abstract

The invention discloses an unsupervised pedestrian re-identification method for enhancing sample data, and belongs to the image retrieval direction in the field of computer vision. An image generationmethod and a feature learning method are combined to solve the domain adaptation problem of unsupervised pedestrian re-identification and keep pedestrian data to be diversified and similar enough. Specifically, a generative adversarial network is utilized to generate sample data with a style similar to that of a target domain to solve the domain style difference problem; meanwhile, in a positiveand negative example selection strategy of a triple loss function, the diversity and similarity of pedestrian data are met by constraining the distance between source domain data and target domain data features, that is, it is ensured that similar samples are sufficiently similar by shortening the distance, and it is ensured that sample data are sufficiently different by shortening the distance. Under the condition that the training cost is not increased basically, the sample data is enhanced, the effect of unsupervised retrieval is greatly improved, the use value is high, and the expandability is high.

Description

technical field [0001] The invention belongs to the field of computer vision, is an important application in the field of image processing, and in particular relates to an unsupervised pedestrian re-identification method for enhancing sample data. [0002] technical background [0003] With the development of deep learning and the improvement of smart devices, high-tech intelligent landing technologies such as intelligent video surveillance, intelligent security, and intelligent transportation are gradually applied to urban life. The pedestrian re-identification technology is mainly to find out the picture information of the pedestrian under other cameras in the short-term scene, extract the pedestrian's external shape, clothing color and other characteristic information through the network, and use the similarity measurement method to match the pedestrian's characteristics according to the pedestrian's characteristics. For the most similar previous images across devices, the...

AI Technical Summary

Problems solved by technology

Since the data sets of the two domains do not overlap, that is, there is no image of the same pedestrian, a part of unlabeled target data is selected to assist the preliminary training process, that is, the images of the two data sets are combined for training, in order to make the network quickly and well adapt to different The distribution of domain data, that is, to solve the problem of unsupervised domain adaptation, use the training data to generate images of various styles using the generative confrontation network. The problem of unbalanced negative examples; the second is to adapt the style of the source domain data to the style of the target domain data through the generated sample data, and constrain the source domain data and the target domain data through the loss function, which not only ensures the problem of domain adaptation, It can also ensure the diversity and similarity between samples to achieve better test results

Images