Vehicle re-identification method based on multi-branch deep learning

Abstract

The invention discloses a vehicle re-identification method based on multiple branches. The vehicle re-identification method comprises the following steps: transmitting a vehicle to-be-identified imagethrough a traffic camera; inputting the to-be-identified image into a training model, obtaining feature vectors corresponding to the plurality of branches, specifically, giving the input vehicle image, an RAM (Random Access Memory) generating a group of function vectors, in particular, generating a feature map M by five shared convolutional layers, then feeding M to four branches to generate different features, wherein the four branches comprise a global branch, a BN branch, an attribute branch and a local area branch, and extracting corresponding feature vectors from the four branches; and comparing the feature vector of the to-be-identified image with the feature vector of the vehicle of the video image in the effective geographic area range, searching the vehicle target image with thehighest similarity by using a similarity calculation formula, and outputting a final re-identification system model. According to the method, the specific target vehicle can be tracked and found, andrelated departments are assisted to improve the target vehicle searching efficiency.

Description

technical field [0001] The invention relates to the technical fields of computer vision and intelligent transportation, in particular to a multi-branch-based vehicle re-identification method. Background technique [0002] With the continuous development of artificial intelligence, computer vision and hardware technology, video image processing technology has been widely used in intelligent transportation systems (ITS). In recent years, with the promotion of highway video surveillance, image processing methods have begun to be applied to the field of traffic analysis, including traffic incident detection, traffic queue monitoring, vehicle type recognition, vehicle classification, traffic flow statistics, etc. [0003] Vehicle re-identification (Re-ID) is an important research direction in the field of computer vision, focusing on the recognition of specific target vehicles under cameras without public view without displaying license plate information. The vehicle re-identifi...

AI Technical Summary

Problems solved by technology

[0004] With the increasing amount of surveillance video data, relevant departments often need to search for massive video data when obtaining the required clues from these surveillance videos. For the driving record in the past month, it is necessary to watch the surveillance video of the major streets and roads in the past month, and use the existing photos of the suspect vehicle to identify the picture where the suspect vehicle is located through human eyes or simple image matching. In this way, the driving trajectory of the suspected vehicle can be summarized. The identification method of this suspected vehicle is too simple, and the identification accuracy is low.

Images