Cross-modal pedestrian re-identification method based on self-imitation mutual distillation

Abstract

The invention discloses a cross-modal pedestrian re-identification method based on self-imitation mutual distillation, and relates to the field of image processing. Aiming at the defect that an existing one-stage feature registration method ignores feature distribution differences in modals and between modals, a two-stage feature registration method is provided, and the performance of cross-modal pedestrian re-identification is improved. The feature registration of the two stages comprises the following steps: 1) feature registration in a mode: obtaining prototype features of each pedestrian category in a self-simulation learning mode, and realizing feature registration in the mode by improving the similarity between all samples of the category and the prototype features; 2) feature registration between modals: adopting a mutual distillation learning method to reduce the distribution difference of samples of the same category and different modals. The discriminability of features is improved. All samples from two different modalities and the same ID learn the feature distribution of each other, so that the feature difference between the modalities is reduced. The method can be used for intelligent video monitoring, pedestrian tracking and behavior analysis, intelligent security and protection and the like.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a cross-modal pedestrian re-identification method based on self-imitation and mutual distillation, which can be used for intelligent video monitoring, pedestrian tracking and behavior analysis, intelligent security and the like. Background technique [0002] Cross-modal person re-identification has received extensive attention in recent years because of its application prospects and practical application value, and many excellent algorithms have emerged. These algorithms can be broadly classified into three categories: feature registration-based cross-modal person re-ID algorithms, image generation-based cross-modal person re-ID algorithms, and metric learning-based cross-modal person re-ID algorithms. Compared with the other two types of algorithms, the cross-modal person re-identification algorithm based on feature registration has received more attention. It achieves the goal ...

AI Technical Summary

Problems solved by technology

In addition to increasing the complexity of the model, this type of method ignores the impact of redundant information within the modal on the accuracy of cross-modal pedestrian retrieval, and only directly performs one-stage feature registration, which cannot effectively alleviate the impact of differences between modalities. Barriers to Performance Improvement

Images