An Accurate Pedestrian Re-identification Method In Cross-View Video Image Considering Small Object Feature

Video surveillance plays an irreplaceable role in safeguarding social security and maintaining public order,so it is widely used in schools,shopping malls and other crowded public places.Video images are used to carry out pedestrian detection and cross-view pedestrian re-identification.It has important application value in the fields of smart city and criminal investigation.However,the existing pedestrian detection and cross-view pedestrian re-identification methods often have problems such as low detection and re-identification accuracy,which makes them unable to meet the technical requirements for practical use.In view of the above problems,this paper conducts research on the accurate pedestrian re-identification method in cross-view video images considering small object feature,selects case scenarios and conducts experimental analysis.The main research work are summarized as follows:(1)An intelligent pedestrian detection method in video images considering small object feature was proposed.Firstly,we give out the general idea of pedestrian intelligent detection.Then we perform the appearance analysis and deep feature visualization of pedestrians in video images to provide strategies for model design.Finally,we propose an improved Faster R-CNN algorithm considering small object feature,construct a multi-scale fusion feature extraction network,and design a structure-aware optimization region proposal network.It solves the problem of false detection and missing detection of small-scale pedestrian,and improves the accuracy of pedestrian detection.(2)An accurate cross-view pedestrian re-identification method with multi-level feature constraints was proposed.Firstly,we analyze the application limitations of traditional pedestrian re-identification methods,and give the general research ideas for accurate pedestrian re-identification.Then we use the precise detection frame coordinates,pedestrian detection frame confidence and pedestrian identity features to constrain the network model with multi-level features.Finally,we design an integrated model of pedestrian detection and pedestrian re-identification with multi-level feature constraints,and give the specific process of the network model to achieve accurate pedestrian re-identification in cross-view video images.(3)An experimental analysis of case scenarios was carried out.Based on the above methods,we constructed a dataset and selected campus video images for experimental analysis.The experimental results show that the m AP of the pedestrian detection method in this paper can reach more than 85%,which is 7.8%-24.5% higher than that of the classical methods,and can achieve accurate detection for small-scale object.The Top-1 index of the pedestrian re-identification method in this paper can reach more than89%,which is 8%-10.7% higher than that of the excellent methods in recent years.It can accurately identify pedestrians with similar appearance and large scale differences.