Vehicle re-identification method and device based on multi-dimensional features

Abstract

The invention discloses a vehicle re-identification method and device based on multi-dimensional features. The method comprises the following steps: acquiring an image frame containing a target vehicle through a monitoring camera; performing vehicle re-identification model identification on the image frame; performing vehicle identification by using a vehicle re-identification model from the imageframe to obtain a first identification domain; identifying a multi-dimensional feature model of the image frame: obtaining multi-dimensional features of the vehicle from the image frame by using themulti-dimensional feature model, encoding the multi-dimensional features, and matching based on the encoded multi-dimensional features to obtain a second identification domain; and solving an intersection of the first identification domain and the second identification domain to obtain a target vehicle set. According to the invention, a vehicle re-identification and multi-dimensional feature combined identification mode is adopted, and the re-identification accuracy can be improved, so that the problems of no license plate and very high vehicle similarity are solved.

Description

technical field [0001] The present application relates to a vehicle re-identification method and device based on multi-dimensional features. Background technique [0002] With the continuous improvement of people's living standards, the requirements for travel are also getting higher and higher, the number of urban motor vehicles is increasing, and the contradiction between vehicle management and effective traffic is becoming increasingly prominent. Vehicle re-identification refers to the process of retrieving vehicle pictures under different surveillance cameras and finding the target vehicle in a large number of vehicle pictures and video clocks. Vehicle re-identification is of great significance for criminal investigation, high-speed toll inspection, public management, smart transportation, etc. Most of the existing re-identification methods based on deep features only use the global features of the entire picture, some salient features or only use models such as twin ne...

AI Technical Summary

Problems solved by technology

This application adopts the combination of vehicle re-identification and multi-dimensional feature recognition, which can increase the accuracy of re-recognition, thereby solving the problem of no license plate and high vehicle similarity. There are license plate recognition errors in multi-dimensional features. For problems such as vehicle model recognition errors, local salient features are used. The vehicle re-identification model is to input the entire image to obtain the global features of the entire image. According to the vehicle re-recognition model, the results will not be focused on specific features such as vehicle models. License plate, so the combination of the two can solve the shortcomings when the two are used alone, greatly improve the accuracy of recognition, and because of the fuzzy recognition, it has the basis of combining the driving route of the relevant vehicle. On this basis, the vehicle can be realized. Accurate Restoration of Driving Route

Images