Vehicle Re-identification Based On Deep Learning

With the increasing number of vehicles in modern society,various traffic problems have followed.In this case,vehicle re-identification has gradually been widely studied.The goal of vehicle re-identification is to identify the same car below the pictures taken by different cameras.Today's focus is on vehicle re-identification in urban surveillance scenarios,so the dataset used in the experiment is a real-time vehicle photo taken by the surveillance camera in the city.At present,the photos of vehicles photographed in real-time under traffic conditions may not be able to capture license plates,or when facing unlicensed cars and deck cars,vehicle identification based on license plate information may not be able to complete the task,so it is necessary to study Re-recognition of vehicles based on license plate information.This is very important for the identification of unlicensed vehicles in the deck car and the judgment of the escape route of the vehicle.Based on this research,the convolutional neural network is used as the feature extractor to replace the traditional HOG or SIFT image feature extraction method,and the validity of the feature extracted by the convolutional neural network is verified.Furthermore,the VGG network is further optimized by fine-tuning,and a new model combining the features extracted by multiple convolutional neural networks is proposed.The color features proposed by the Double AlexNet model and the advanced semantic features extracted by GoogLeNet merged with the features extracted by the VGG network successfully improved the Rank-1 and Rank-5.the images searched by the data set are sorted and output according to the distance of the Euclidean distance.the optimization of the deep learning model is verified by the proposed feature fusion model.