Cross-modal pedestrian re-identification method based on dynamic redundancy suppression

Abstract

The application discloses a cross-modal pedestrian re-identification method based on dynamic redundancy suppression, comprising the following steps: preprocessing visible light pedestrian images and infrared pedestrian images of a search target and a matching database; inputting the images into an optimized dynamic redundancy suppression cross-modal model to obtain identification features; obtaining an identification matching result according to the similarity of the identification features of the search target and the matching database; the dynamic redundancy suppression cross-modal model firstly performs convolution processing on the input images to obtain visible light features and infrared features, calculates channel attention information and fuses the features through two intra-modal feature learners and one cross-modal feature learner to obtain auxiliary modal features; and then the features are respectively input into weight-shared feature encoders, global average pooling and batch normalization are performed to obtain the identification features. The application effectively alleviates the intra- and inter-differences between visible light and infrared modalities, and fully mines the identity features of pedestrians.

Images