End-to-end pedestrian re-identification method based on self-attention deep learning

Abstract

The invention discloses an end-to-end pedestrian re-identification method based on self-attention deep learning, relates to the technical field of machine learning and pattern identification, and is used for improving pedestrian re-identification performance. The method comprises the following steps: 1) taking a triple image as input data of a deep learning network, and removing the last classification layer by adopting a pre-trained Resnet50 deep network to extract depth features of the image; 2) on the basis of the depth features, further obtaining self-attention features through a self-attention network; (3) fusing the self-attention feature and the depth feature to generate an image feature with higher identification capability, and (4) jointly supervising the training of a network byloss functions of multiple classification tasks and verification tasks, and continuously optimizing network model parameters through multiple iterations to obtain an optimal model for pedestrian re-identification.

Description

technical field [0001] The invention relates to the technical field of machine learning and pattern recognition, in particular to an end-to-end pedestrian re-identification method based on self-attention deep learning. Background technique [0002] Pedestrian re-identification, as an important intelligent video analysis technology, has important research value for cross-camera pedestrian target tracking and pedestrian behavior analysis. Pedestrian re-identification is to identify images of pedestrians with the same identity across multiple cameras. Cross-camera pedestrian images often face complex background clutter, illumination changes, severe occlusion, significant pose changes, etc., so the research on pedestrian re-identification is extremely challenging. [0003] Extracting more discriminative and robust features from original pedestrian images is one of the important research tasks of pedestrian re-identification. With the development of deep learning technology, de...

AI Technical Summary

Problems solved by technology

However, changes in camera perspective and pedestrian pose can cause uncontrollable misalignment between cross-camera pedestrian images, making it difficult to learn a more discriminative and robust feature model to cope with complex scene changes across cameras

Images