Research On Person Re-identification Algorithm Based On Attention Mechanism And Attribute Learning

In recent years,with public safety issues have received more and more attention,surveillance systems have been widely used in all corners of society,and therefore a large amount of data has been generated.It is necessary to quickly search and track target pedestrians in numerous video data.It seems increasingly difficult.In order to reduce the workload of surveillance personnel and increase search efficiency,the method of Person re-identification(re-ID)has been regarded as one of the hotspots of current scholars.Person re-identification refers to an algorithm that can retrieve target pedestrians from multiple non-overlapping cameras.The Person re-identification method originally uses computer image processing technology to extract the histogram features of the pedestrian,and then uses the distance measure to calculate the similarity of the pedestrian.Due to the complexity of feature extraction and the low recognition rate,this method gradually fails to satisfy people's requirements.With more and more applications and breakthroughs made by convolutional neural networks in the field of image recognition in recent years,the application of convolutional neural networks to Person re-identification algorithms has gradually become one of the current hot research topics.However,there are still many challenges in the field of Person re-identification.As the cameras in the surveillance system are distributed in different areas,the camera parameters are uneven,and the behavior and attitude of pedestrians are constantly changing,which leads to the inconsistency of the performance of the same pedestrian in different cameras.In addition,the target pedestrian will be blocked by other pedestrians or objects in the camera picture,which also greatly improves the difficulty of pedestrians in the algorithm of gender recognition.Aiming at the problems in Person re-identification,this paper designs an end-to-end Person re-identification model.The model uses a convolutional neural network to extract global features of pedestrians.By introducing attention mechanism to align pedestrian parts and extract fine-grained features on pedestrian parts,the accuracy of person re-identification can be greatly improved.At the same time,attribute information with semantic invariance is introduced,one-to-one correspondence between attribute information and pedestrian parts,and a corresponding loss function is designed.By optimizing training,the model learns the attributes of pedestrian parts,so that the attention mechanism can be more accurate.Locate the pedestrian.Finally,the Person re-identification algorithm based on attention mechanism and attribute learningproposed in this paper has achieved high accuracy through experimental simulation on public data sets.Compared with the advanced algorithms in recent years,especially in dealing with the problems of pedestrian misalignment and occlusion,it shows good performance,which proves the competitiveness of the algorithm in this paper.