A pedestrian retrieval method based on augmented samples and multi-stream layers

Abstract

The embodiment of the present invention discloses a pedestrian retrieval method based on augmented samples and multi-stream layers. The method includes: constructing a deep neural network based on augmented samples and multi-stream layers; obtaining a training set, and using deep convolution to generate an adversarial network Generate generated samples to expand the training set; select B real samples and B generated samples from the training set as the input of the deep neural network; use the multi-flow layer of the deep neural network to obtain the pedestrian feature representation; send it to the mixed quadruple loss, and based on the loss value Optimize the deep neural network; use the trained deep neural network to extract the final feature representation of the sample to be queried, and use the similarity between feature vectors to obtain the matching result of the sample to be queried. The present invention uses multi-flow layers to learn pedestrian features in different aspects, and uses mixed quadruple loss to obtain a discriminative feature space, which not only reduces the risk of over-fitting but also improves the generalization ability of the network, thereby improving pedestrian retrieval matching accuracy.

Description

technical field [0001] The invention belongs to the technical fields of computer vision, deep learning and artificial intelligence, and in particular relates to a pedestrian retrieval method based on augmented samples and multi-flow layers. Background technique [0002] Pedestrian retrieval aims to retrieve specific pedestrians from cameras set up at different angles, and it plays an important role in applications such as video retrieval, multi-camera tracking, and behavior recognition. However, pedestrian retrieval still faces many challenges, such as pose changes, viewpoint changes, illumination changes, and occlusions. [0003] Currently, due to the advantages of deep neural network in feature learning, it is widely used in the field of pedestrian retrieval. However, the number of training samples and model structure have a large impact on the performance of deep neural networks. First, when training deep neural networks, a large number of samples are required, but most...

AI Technical Summary

Problems solved by technology

However, the number of training samples and model structure have a large impact on the performance of deep neural networks

First, when training deep neural networks, a large number of samples are required, but most existing databases are relatively small, which may lead to overfitting

Second, in practical applications, since pedestrians may wear similar clothes, it is difficult to distinguish them from their appearance

However, the structure of these network models is relatively simple, and it cannot learn the feature representation of pedestrians from many aspects.

Images