Text-based vehicle image fine-grained retrieval system

Abstract

The invention provides a text-based vehicle image fine-grained retrieval system, which allows a user to acquire required information in a natural language retrieval mode, and reduces a retrieval threshold compared with a method for retrieving an image by using an image. A cross-modal semantic matching task between the text and the image is converted into a text semantic matching task which is moredirect, high in interpretability, simple and mature in research; the retrieval text of the near-end time is periodically analyzed, and interest words and interestingness are extracted to establish aninterest word list, and the interest word list is utilized to predict retrieval trends of users, and fine-grained description texts which better accord with recent user interest points are generatedfor images, and dynamic updating of an image description text library is realized, so that the flexibility and retrieval performance of the method are improved; and fine-grained characteristics of each part of the vehicle can be accurately captured, so that different vehicles of the same vehicle type can be identified more accurately.

Description

technical field [0001] The invention relates to the field of traffic image processing, in particular to a text-based fine-grained vehicle image retrieval system. Background technique [0002] At present, cases of stealing vehicles and using vehicles to commit crimes have increased repeatedly, seriously affecting social stability. In order to evade police investigation, suspects often change vehicle license plates, remove or block vehicle locators and other behaviors. In this case, it is necessary to rely on vehicle appearance features to track and locate the target vehicle. Vehicle image retrieval is an important part of the intelligent transportation system. It retrieves images of vehicles whose appearance features match the retrieval conditions from the images captured by road monitoring to assist the police in tracking stolen, hit-and-run or other criminal-related vehicles. Measuring the semantic similarity between an image to be retrieved and a retrieval condition is a...

AI Technical Summary

Problems solved by technology

However, in actual situations, it is usually difficult to construct an image that meets the retrieval conditions, which limits the application scenarios of this type of method.

In addition, there are also some works to retrieve images by automatically generating text indexes, but the generated text lacks descriptions of fine-grained features of images

Images