A car search method based on deep learning

Abstract

The invention relates to the technical field of computer vision, specifically a deep learning-based image search method, which integrates global appearance features and local features of vehicles, and has strong generalization and robustness. Including steps: collecting pictures of actual application scenarios; automatically labeling the acquired pictures, that is, the pictures of the same car are classified into one category, establishing a training set and a test set; designing the neural network structure; inputting the samples of the training set into the neural network Carry out training in the structure to obtain the image search model; use the image search model to calculate the feature vectors of all samples in the test set; feature comparison: use the cosine distance to calculate the similarity between the target image feature vector and the sample feature vector; The obtained similarities are sorted from largest to smallest, and the results of car search by image are obtained.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a method for searching cars by image based on deep learning. Background technique [0002] Searching cars with pictures is an image-based vehicle retrieval technology, which aims to search for the same vehicle in different video surveillance scenarios. Statistics show that more than 65% of crimes are related to vehicles. In car-related cases, the license plate of the suspect vehicle is usually a fake or fake plate, and the identity of the vehicle cannot be locked by identifying the license plate number. Therefore, the technology of searching for cars by image based on visual appearance features has great research significance and practical value. [0003] At present, the image search technology is mainly divided into two categories, one is based on traditional image features, and the other is based on deep features. Based on traditional image features, manually designed...

AI Technical Summary

Problems solved by technology

This method is only applicable to the front shot of the vehicle, and when the pasted signs are changed, the accuracy of vehicle recognition will be significantly reduced

Images