Vehicle re-identification method based on multi-attribute deep features

Abstract

The invention discloses a vehicle re-identification method based on multi-attribute depth features, comprising: using a feature extraction model to extract the depth features of a test picture set of the A-th pooling layer, using the depth features of the test picture set and a W matrix to obtain Find the Mahalanobis distance between the depth feature of the test picture in the set and the depth feature of the target picture in the candidate set, sort according to the Mahalanobis distance from small to large, and obtain the similarity sorting result of the test picture and the target picture; the test picture set Comprising a lookup set and a search set, the test picture set is a picture that includes a vehicle; the training of the feature extraction model includes: accessing the vehicle multi-attribute classifier after the A pooling layer of GoogLeNet to obtain improved GoogLeNet, using Training pictures are trained to improve GoogLeNet and obtain a feature extraction model. The invention simplifies the model training process, greatly improves the accuracy of re-identification, and has strong model generalization performance.

Description

technical field [0001] The invention belongs to the technical field of deep learning, and more specifically relates to a vehicle re-identification method based on multi-attribute deep features. Background technique [0002] Vehicle re-identification is an important research direction in the field of computer vision, focusing on the recognition of specific target vehicles without license plate information under cameras without public view. As a newly emerging research field, although vehicle re-identification is of great significance to intelligent transportation and other aspects, there are still few related studies. One of the current mainstream methods is to use an end-to-end deep neural network to train a convolutional neural network through multiple sets of image pairs of the same target and different targets, and simultaneously pursue the minimization of the intra-class distance and the maximization of the inter-class distance during training. Another method is to find...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the above defects or improvement needs of the prior art, the present invention provides a vehicle re-identification method based on multi-attribute deep features, thereby solving the technical problems of the prior art such as complicated training process, poor re-identification effect, and low accuracy rate

Images