Data-driven pedestrian re-identification based on hierarchical semantic representation

摘要

Limited number of labeled data of surveillance video causes the training of supervised modelfor pedestrian re-identification to be a difficult task. Besides, applications of pedestrianre-identification in pedestrian retrieving and criminal tracking are limited because of the lack ofsemantic representation. In this paper, a data-driven pedestrian re-identificationmodel based onhierarchical semantic representation is proposed, extracting essential featureswith unsuperviseddeep learning model and enhancing the semantic representation of features with hierarchicalmid-level ‘attributes’. Firstly, CNNs, well-trained with the training process of CAEs, is used toextract features of horizontal blocks segmented from unlabeled pedestrian images. Then, thesefeatures are input into corresponding attribute classifiers to judge whether the pedestrian hasthe attributes. Lastly, with a table of ‘attributes-classes mapping relations’, final result can be calculated.Under the premise of improving the accuracy of attribute classifier, our qualitative resultsshow its clear advantages over the CHUK02, VIPeR, and i-LIDS data set. Our proposed methodis proved to effectively solve the problem of dependency on labeled data and lack of semanticexpression, and it also significantly outperforms the state-of-the-art in terms of accuracy andsemanteme.