Cross-resolution vehicle re-identification method based on attention guidance generative learning

Abstract

The invention relates to a cross-resolution vehicle re-identification method based on attention guidance generative learning. The method comprises the following steps of: constructing a vehicle imagesuper-resolution network based on generative learning, an attention guidance mechanism, an end-to-end vehicle image super-resolution model and re-identification model joint mechanism, and adopting a loss function, thereby realizing efficient cross-resolution vehicle re-identification. According to the method, the problem that the cross-resolution vehicle image recognition precision is not high invehicle re-recognition is particularly considered, that is to say, a query image captured in the actual situation is usually a low-resolution image and cannot be accurately matched with a high-resolution candidate image in a query library, and the cross-resolution vehicle re-recognition precision can be effectively improved.

Description

technical field [0001] The invention relates to the fields of computer vision and pattern recognition, in particular to a cross-resolution vehicle re-identification method based on attention-guided generative learning. Background technique [0002] Vehicle re-identification aims at matching images of the same vehicle captured by different cameras, which has very important practical application value. Vehicle re-identification has a wide range of application scenarios, such as vehicle tracking, vehicle positioning, criminal detection, etc., and is an important part of the intelligent monitoring system. [0003] Among many factors such as illumination changes, viewpoint changes, occlusions, and resolution changes, resolution changes (i.e., different resolutions of vehicle images captured by cameras at different distances) are challenging factors in vehicle re-identification. In particular, cross-resolution vehicle image queries are ubiquitous in practice, that is, it is diffi...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that cross-resolution vehicle re-identification is difficult to accurately identify, the present invention proposes a cross-resolution vehicle re-identification method based on attention-guided generative learning, constructs a vehicle image super-resolution network based on generative learning, and an attention-guided mechanism. The end-to-end joint mechanism of vehicle image super-resolution model and re-identification model proposes a joint loss function, which effectively improves the cross-resolution recognition accuracy of vehicle re-identification

Images