Pedestrian re-identification measurement method based on triangular model

Abstract

The invention discloses a pedestrian re-identification measurement method based on a triangular model, and belongs to the technical field of computer vision and mode identification. According to the invention, fine-grained features of pedestrian images are extracted through a BFE model, and the distance between two sample points is determined through a triple loss function for measurement. According to the invention, a loss function is utilized to have good supervision performance on training of a plurality of model structures, pedestrian re-identification accuracy can be effectively improved,and the invention has good generalization ability on three data sets. According to the invention, a BFE model structure is adopted, the triple loss function method based on the triangular model can effectively improve pedestrian re-identification accuracy, and a loss function has a stronger supervision effect on model training.

Description

technical field [0001] The invention belongs to the technical field of computer vision and pattern recognition, and in particular relates to a pedestrian re-identification measurement method based on a triangular model. Background technique [0002] In person re-identification research, feature extraction and similarity measurement are important steps in the re-identification task. Although the deep neural network can extract the deep features in the image data, so that the performance of the model can be improved, but the appropriate loss function is also the core issue to improve the re-identification performance. A good similarity measure can improve the search performance between images and improve the model re-identification accuracy, especially when the number of pedestrian categories is large or unknown. The triplet loss function is widely used in person re-identification tasks. In the present invention, we build a triangular model for anchor samples, positive sampl...

AI Technical Summary

Problems solved by technology

Although the deep neural network can extract the deep features in the image data and improve the performance of the model, the appropriate loss function is also the core issue to improve the re-identification performance.

Images