Research Of Person Re-identification Based On Self-Attention And Image Dicing Multi-feature Fusion

Person Re-Identification(Re-ID) is a technology that uses computer vision technology to determine whether there is a specific pedestrian in an image or video sequence.It is often used in urban security,video surveillance and other fields.With the development of computer vision and deep learning technology,the performance of person re-identification has been greatly improved.However,there are still some problems in practical applications,such as camera viewing angle differences,lowresolution images,and similar appearance of pedestrians.In order to solve the problem of similar appearance of pedestrians in the process of person Re-Identification,this thesis explores the person re-identification method based on the self-attention mechanism and the person re-identification method based on image dicing.The main research contents and innovations are:(1)Aiming at the problems that the ordinary convolutional neural network model cannot distinguish the importance of different location features,and it is difficult to obtain distinguishable important pedestrian features,a person re-identification method based on the self-attention mechanism is proposed.This method obtains the relationship between any two pixels in the image by introducing the self-attention mechanism module,thereby paying more attention to representative pedestrian characteristics in the feature extraction process,making the distinguishing characteristics between different pedestrians more obvious,and improving person the effect of re-identification.In order to show the effectiveness of the proposed method,this thesis conducts comparative experiments on mainstream data sets.The experimental results show that the accuracy of the proposed method is better than other methods based on global feature extraction.(2)Aiming at the fact that a single global feature ignores many local features and makes it difficult to capture distinguishing pedestrian features,while a single local feature cannot take into account the overall feature problem,a multi-feature fusion person re-identification model based on image dicing is proposed.This method uses Based on image dicing and multi-feature fusion technology,the network can learn not only the global features of pedestrians at different levels,but also the distinguishing detailed features.The model extracts 3 branch features respectively.Branches 1 and 2are responsible for extracting global features of different levels of pedestrians.Branch3 uses horizontal block technology to divide the image level into upper and lower parts and perform local feature extraction respectively,and finally uses the multi-feature fusion technology deeply integrates the characteristics of the three branches.Aiming at the difference between global features and local features,three kinds of loss functions are combined for supervised training.The proposed model has been validated and tested on three mainstream data sets.The experimental results show that compared with other methods,this model has achieved better experimental results on both Rank-1 and m AP indicators,indicating the effectiveness of the method.