Vehicle re-identification method based on deep learning

Abstract

The invention discloses a vehicle re-identification method based on deep learning, and the method comprises the following steps: employing a batch-difficult triple loss to train a convolutional neuralnetwork, and employing a triple to capture the relative similarity among three elements in the training process to learn representative features; cross-merchant loss is adopted for training, and labels are added on the basis for smoothing and standardization; carrying out joint training on the cross-quotient loss of the batch-difficult triple loss and the label-added smooth normalized loss; and the result is evaluated on the reference data set through experiments. According to the method, the problems that in the prior art, a plurality of triplets continuously go deep into a useless triplet along with training, and training is time-consuming are effectively solved, and the problem that the generalization ability and adaptability of the model are reduced due to training overfitting can beweakened through label smoothness standardization.

Description

technical field [0001] The invention belongs to computer vision technology, in particular to a vehicle re-identification method based on deep learning. Background technique [0002] As video surveillance occupies an increasingly important position in the field of public safety, more and more attention has been paid to vehicle-related tasks, such as vehicle detection, tracking, classification, and verification. It has become an urgent need to apply object re-identification technology to the field of vehicle recognition. How to troubleshoot the target vehicles that the public security department needs to track in complex urban video images is particularly important. The task of vehicle re-identification is to search a database for images of the same vehicle captured by multiple cameras in non-overlapping regions, given a known target vehicle image. Different from vehicle detection, tracking and classification, vehicle re-identification can be treated as an instance-level obj...

AI Technical Summary

Problems solved by technology

[0005] However, compared with pedestrian images, different vehicle images have more similar appearance information, and the same vehicles may have relatively large differences in appearance due to factors such as weather and illumination, so vehicle re-identification has more many challenges

In addition, there is a special difficulty in vehicle re-identification. Since vehicles belonging to the same model are very similar to each other, it is very difficult to distinguish vehicles with different IDs but belonging to the same model with human eyes.

Images