Domain adaptive pedestrian re-identification method based on style conversion and joint learning network

Abstract

The invention relates to a domain adaptive pedestrian re-identification method based on style conversion and a joint learning network. The method comprises the following steps: 1, pre-training a neural network model by using a source domain data set; 2, performing style conversion on the pedestrian image in the target domain data set; 3, preprocessing each pedestrian image; 4, inputting the same pedestrian image obtained by adopting two preprocessing modes into two neural network models to extract features, and storing two high-order features into two storages; clustering the two high-order features to obtain a pseudo tag; fusing the two high-order features of the same pedestrian image, and storing the fused high-order features in a joint memory; 5, training the two neural network models based on the pseudo tag, and synchronously training the two neural network models based on the joint memory; and 6, repeating the step 4 and the step 5, calculating the recognition precision of the two neural network models in the training process, and using the neural network model with the optimal recognition precision for pedestrian re-recognition.

Description

technical field [0001] The invention belongs to the technical field of image data processing in computer vision, in particular to a domain adaptive pedestrian re-identification method based on style conversion and joint learning network. Background technique [0002] Pedestrian re-identification is a technology to judge whether there is a specific pedestrian between different cameras across time and space. This technology is more inclined to the overall consistency in the mining of pedestrian characteristics, and is not limited to whether pedestrians are dressed in the same way or whether their faces are blocked. and other representative details. At present, this technology is widely used in many fields such as intelligent security, unmanned supermarkets, and human-computer interaction. Pedestrian re-identification has important research value and research needs in both academia and industry. How to accurately match people under different cameras? The same pedestrian identi...

AI Technical Summary

Problems solved by technology

The Chinese patent with the publication number CN111898665A discloses a domain-adaptive pedestrian re-identification method guided by neighbor sample information. The domain transitions to the target domain. Because the model relies too much on the influence of neighbor information on clustering, the clustering noise will gradually expand with training.

Images