Research On Key Technologies Of Vehicle Re-identification In Smart City

With the continuous development of computer vision and artificial intelligence technology,people pay more and more attention to the construction of smart city and intelligent transportation.As one of the key technologies,vehicle re-identification has gradually become one of the research hotspots in academia.In this paper,three key problems of vehicle re-identification based on image,sequence and unsupervised cross domain adaptive are studied.The main research contents and achievements are as follows:1.In vehicle re-identification based on image data,a dual branch network structure is proposed,which integrates spatial attention and channel attention.It can effectively capture the salient features of image data in spatial and channel dimensions,and improve the representation ability of single frame vehicle features;2.In vehicle re-identification based on sequence data,an inter frame information mining method and a multi frame information fusion method based on attention are proposed,which can effectively reduce the inter frame information redundancy,mine the complementary features between frames,and form a more comprehensive vehicle feature representation.At the same time,through the efficient fusion of multi frame information,the joint extraction of vehicle spatiotemporal features is realized;3.In unsupervised cross domain adaptive vehicle re-identification,a progressive domain adaptive learning framework based on teacher-student model is proposed.By introducing source domain data supervision and improved triple loss,the influence of clustering noise is reduced,and the cross domain adaptive performance of vehicle re-identification is improved.4.The effectiveness of the three algorithms is verified by experiments on several public datasets and comparison with the current advanced algorithms.