Semantic-Based Deep Learning Algorithm for Vehicle Re-identification.

Re-identification of vehicles is very important in traffic safety, intelligent transportation, and smart cities. The traditional method of identifying vehicles is through license number. However, in some cases, the license number may be blocked, damaged, or fake. In this case, re-identification of vehicles is a difficult challenge without license number. In this article, we propose a vehicle re-identification method based on vehicle semantic information and deep learning. First, extract the overall feature value of the vehicle and use it to identify vehicles with different shapes. Secondly, the semantic information of the vehicle image is perceived, and the characteristic value of the vehicle semantic information is extracted based on the above results, which is used to identify vehicles with the same or similar appearance. Then the overall feature value of the vehicle image and the feature value of semantic information are fused into a comprehensive feature value. Use the generated feature value to calculate the distance to the feature value of other vehicle images, and re-identify massive vehicle images. Finally, we conducted verification on two different data sets. The experimental results show that the proposed algorithm is capable of producing better results.