Vehicle re-identification method based on deep learning

Abstract

The invention discloses a vehicle re-identification method based on deep learning, and belongs to the technical field of image identification and deep learning. The method comprises the following steps: firstly, extracting color features in an HSV space and extracting multi-dimensional features of an RGB image from a target vehicle picture and collected vehicle pictures, and then cascading the target vehicle picture and the collected vehicle pictures to obtain final features, thereby screening out pictures of the same category as the target vehicle picture from the collected vehicle pictures; then, feature extraction is carried out on a target vehicle picture and the screened pictures to obtain high-dimensional features, similar DC module processing and difference DC module processing are carried out on the features on the basis of the features of the same category in a feature set, and the two processed features are fused into a new feature; and finally, judging whether the new features and the target vehicle picture are the same vehicle or not according to the new features. Experiments prove that the method can improve the accuracy and speed of re-identification.

Description

technical field [0001] The invention relates to a vehicle re-identification method based on deep learning, which belongs to the technical fields of image recognition and deep learning. Background technique [0002] Vehicle re-identification refers to the technology of finding the same vehicle that appears in different space-time positions based on the vehicle information collected by the fixed position sensor. Vehicle re-identification technology is one of the key technologies in the smart city system. In order to strengthen road traffic management, the monitoring coverage rate of urban road traffic is getting higher and higher, and more and more video image data are generated every day. When the amount of video data reaches a certain scale, it is obviously unrealistic to rely on manual analysis and processing of these data. With more and more video data, the use of human resources for monitoring and screening has become beyond our means. Using computer vision-related tech...

AI Technical Summary

Problems solved by technology

[0005] In view of the above problems, the present invention proposes a vehicle classification and re-identification method based on deep learning, which is classified in two steps. First, a rough classification is performed, and unnecessary vehicles that are easily distinguished with large differences are eliminated, and then the DC module is used for re-identification. Recognition solves the technical problem of slow speed, partially solves the problems of inaccurate classification and similarity between classes in the existing methods, and provides better auxiliary functions for smart cities

Images