Research On Person Re-identification In Multiplt Feature Fusion

Pedestrian re-identification needs to retrieve specific pedestrians under different cameras in different areas.As the continuous development of the country,the demand for surveillance devices is more and more urgent.Large-scale surveillance provides massive data for the retrieval,tracking and identification of target pedestrians.In actual monitoring scenarios,it is very difficult to perform intelligent recognition only by relying on face,gait and other information.Therefore,more and more scholars have begun to focus on the task of Pedson Re-Rdentification,relying on Person Re-Identification technology to screen and mine from massive data.It can generate effective information,free people from tedious work,and realize cross-camera detection.Due to the small scale of existing datasets,camera shooting angle changes,pedestrian pose changes and other reasons,the Peson Re-Identification task has very important research significance and faces many challenges.It is foreseeable that it will still be a research hotspot in the future,but there are still many areas for improvement.This paper studies the Person Re-Identification method in Multiplt Frature Fusion,and integrates multi-granularity features through network architectures such as convolutional neural networks and Trans Former to improve the accuracy of recognition.The research on Person Re-Identification in Multiplt Frature Fusion is mainly to solve difficult problems such as pedestrian feature alignment,perspective change,and occlusion.This paper mainly completes the following work:(1)For the purpose of capturing the multi-granularity features of pedestrian images and improving the recognition accuracy,a Multi-granularity Feature Fusion Network for Person Re-Identification(MFN)is proposed based on the Omist-Scale Network(OSNet).The MFN network is composed of a global branch,a feature dropout branch and a local branch.The feature dropout branch consists of a Dual-channel Attention Dropout Model,which includes a Channel Attention-based Dropout Moudle(CDM)and a Spatial Attention-based Dropout Moudle(SDM).CDM sorts the attention intensity and dropouts low attention channels,and SDM dropouts the most discriminative features with a certain probability in the spatial dimension.The global branch uses the feature pyramid structure to extract multi-scale features,and the local branch employs a uniform partition strategy but produces only a single identity-prediction loss to extracts key local information.Experiments on three large scale datasets show the effectiveness of MFN.On the Market1501,Duke MTMC-re ID and CUHK03-Labeled(Detected)datasets,m AP/Rank-1 of MFN reaches 90.1%/85.7%,81.8%/91.4% and 80.7%/82.3%(77.9%/80.4%).(2)For the purpose of aggregating the multi-granularity features of pedestrian images and further solving the problem of deep feature mapping correlation,Person Re-Identification based on CNN and Trans Former Multi-scale Learning(CTM)is proposed.The CTM network is composed of a global branch,a deep aggregation branch and a feature pyramid branch.Global branch extracts global features of pedestrian images,and extracts hierarchical features with different scales.The deep aggregation branch recursively aggregates the hierarchical features of CNN and extracts multi-scale features.Meanwhile,The feature pyramid branch is consist of a novel two-way pyramid structure,along with the attention module and orthogonal regularization operation,it can significantly improve the performance of the network.Experiments on three large scale datasets show the effectiveness of CTM.On the Market1501,Duke MTMC-re ID and MSMT17 datasets,m AP/Rank-1 reached 90.2%/96.0%,82.3%/91.6%and 63.2%/83.7%.(3)Based on the pedestrian re-identification algorithm,an intelligent monitoring and tracing system,that is,a pedestrian re-identification system,is established.Based on the MFN and CTM network models mentioned above,a Person Re-Identification System is packaged and constructed.Its main functions include pedestrian detection,pedestrian attribute recognition,and pedestrian re-identification.Pedestrian detection uses the YOLOv3 algorithm to retrieve the pedestrians in the surveillance video,uses a rectangular frame to segment and save it to the image library and marks the confidence level;the pedestrian attribute recognition uses the APR(Attribute-Person Recognition)algorithm to quickly narrow the range of the pedestrian attribute image library through the pedestrian attribute features,which can Quickly select a specific target pedestrian;Person Re-Identification takes the query as the search target,calculates the metric function between the features of the query and the pedestrian features in the gallery,and matches them one by one.Finally,use Py Qt5(Qt Designer)to design the graphical interface of the system to visualize the monitoring video and running results.The display results show that the system has high detection and recognition accuracy and has certain application value in practical scenarios.