For example,Бобцов

IMAGE-BASED APPROACH FOR VEHICLE MODEL RE-IDENTIFICATION

Annotation

Subject of Research. The paper presents a study of the existing methods for identifying and comparing the features of objects used in the re-identification task of vehicle model by its image. This task is one of the most important tasks facing automated traffic control systems, and it is solved by comparing the features of the vehicle being verified with a certain set of features obtained earlier by the monitoring system. Then decision is made whether the compared samples belong to the same vehicle model or to different ones. A method is proposed for feature vectors extraction and comparison of vehicle model according to its image. The method is based on the use of convolutional neural networks. The proposed approach is compared with existing algorithms for vehicle model re-identification by the accuracy criterion. Method. The paper describes the approach for vehicle image feature vector extraction and its subsequent comparison with the reference vector for similarity examination. The approach is based on the method of feature vector extraction, using classification convolutional neural network, and on comparison criterion for feature vectors applying the estimate of coincidental features. Main Results. The proposed vehicle model verification method demonstrates accuracy comparable to modern analogous in scenarios when the testing data have characteristics that coincide with training ones (similar camera model and camera angles are used; the level of lighting and noise are similar; models of re-identifiable vehicles are contained in the dataset used for the classification network training). In case of data significantly different from the training dataset, the method shows a lower computational complexity and uses smaller size of used feature vector and demonstrates significantly higher relative accuracy of re-identification. Practical Relevance. The proposed approach is practically applicable in vehicle identification task for highly loaded traffic control systems.

Keywords

Постоянный URL

Articles in current issue

CONTENT © ITMO UNIVERSITY 1993-2024