An occlusion robust person re-identification method based on global and local feature fusion

Abstract

An occlusion robust pedestrian re-identification method based on fusion of global and local features, involving computer vision technology. Including the following steps: 1) preparation of training data; 2) model design and training; model includes ResNet-50 skeleton network, global branch, local branch and semantic branch, global branch uses SPC loss to extract global features, local branch extracts local features, The semantic branch predicts human semantic labels, and the three branches can be combined for end-to-end training. 3) Use the trained model to extract the global features and local features of all pedestrian images in the training set and test set of pedestrian re-identification data, and predict the semantic labels of pedestrian images, and calculate the non-occlusion area indicator. 4) Calculate the similarity between each pedestrian image in the query set and all pedestrian images in the database, and sort them according to the similarity from large to small, so as to complete pedestrian re-identification. Significantly improved the performance of recognition.

Description

technical field [0001] The present invention relates to computer vision technology, in particular to a kind of global and local feature fusion occlusion robust pedestrian re-identification based on semantic segmentation, which uses semantic segmentation of human body as an auxiliary task to solve the occlusion problem that may occur in real pedestrian re-identification scenarios. recognition methods. Background technique [0002] Person Re-identification (Person Re-identification) refers to retrieving pedestrian images with the same identity as a given query image in a pedestrian image database under the scene of multiple non-overlapping cameras. Pedestrian re-identification can be widely used in intelligent security and video surveillance and other fields. [0003] At present, with the development of deep learning, the performance of the person re-identification problem has been greatly improved. On the one hand, in order to learn more discriminative feature representatio...

AI Technical Summary

Problems solved by technology

Therefore, the occluded person re-identification problem is a more challenging problem

When the detection of key points is inaccurate due to occlusion, the extraction of global features is affected, resulting in a decrease in performance

Images