Pedestrian re-identification method based on global-local feature dynamic alignment

Abstract

The invention discloses a pedestrian re-identification method based on global-local feature dynamic alignment, and the method comprises the steps: carrying out the preprocessing of a training set in a selected pedestrian re-identification data set, and carrying out the initialization of a model through a pre-trained model ResNet50; then, calculating a global distance for the coarse-grained global features and calculating a local alignment distance for the fine-grained local features, and using difficult samples to mine triple loss, center loss and Softmax cross entropy loss to jointly serve as a supervisor of the pedestrian re-identification network, and constraining training of the model; and finally, fusing the global features and the local features to obtain final features, taking the queried pedestrian image as the input of a pedestrian re-identification network model, and retrieving the pedestrian with the shortest alignment distance from the candidate library. According to the method, noise interference of space misalignment and non-alignment areas can be effectively suppressed, additional auxiliary attitude information does not need to be introduced, local branches are used for guiding calculation of distance measurement, and the accuracy of the test stage can be further improved.

Description

technical field [0001] The invention relates to the field of computer, pedestrian re-identification and intelligent monitoring, in particular to a pedestrian re-identification method based on global-local feature dynamic alignment. Background technique [0002] Person Re-Identification (Re-ID) is a challenging task in the field of computer vision, aiming to determine whether pedestrian images captured by different cameras or different video clips of the same camera are the same pedestrian, and it has been widely used in intelligent surveillance field. However, due to the complexity of real-world scenarios, pedestrian Re-ID still faces many challenges, such as pedestrian detection bounding box errors, pose changes, and occlusions. These challenges make it a difficult task to identify specific pedestrians from a large test set. [0003] To address these challenges, most of the previous works have focused on learning global features of pedestrians using Convolutional Neural N...

AI Technical Summary

Problems solved by technology

In general, these methods either need to introduce the pedestrian's pose to assist the alignment of the pedestrian's local features, but need to add additional computing resources; or use a local hard alignment method to match the pedestrian's local features, but encounter pedestrian pose changes, When there are large scene differences such as pedestrian detection bounding box errors and partial occlusions, it is difficult to obtain high pedestrian Re-ID accuracy with hard alignment methods

Images