Person Re-Identification In Printing Intelligent Factory Under Field Of View

In order to realize the intelligent management of employees in printing intelligent factory,an intelligent person re-identification system is needed,person re-identification is to match the target object with other pedestrians in a multi-camera network.person re-identification enables us to find the same object of interest in multiple cameras,which has wide application value in intelligent factories,person re-identification is still a challenging subject due to the influence of various uncertainties.In surveillance video,the significant visual changes of pedestrians under different body postures,different illumination conditions and different viewing angles make the changes of pedestrians themselves greater than those between pedestrians.In addition,background clutter and occlusion can cause additional difficulties.This subject is entrusted by Shaanxi Beiren Printing Machinery Co.,Ltd.to study the problem of fast extraction and accurate identification of pedestrian information in printing intelligent factory.The following work has been carried out in this paper:(1)This paper proposes a person re-identification method based on weighted distance measurement of multiple features.Firstly,human body partition is used to extract multiple local features,feature vectors based on multiple features are established,and a weighted distance measurement method is given.Secondly,a ranking optimization framework is proposed based on the similarity be:tween training set and test set images.In the experiment,this paper validates the effectiveness of the proposed method on iLIDS and VIPeR datasets,and then validates the rank optimization effect on the above two datasets.The results show that the proposed method has strong robustness to perspective,attitude and so on.The recognition rate of rank-1 on iLIDS datasets is about 5%-7%higher than that of other methods.Rank optimization can improve the reco.gnition rate by about 5%.Rank-1 recognition rate on VIPeR datasets is about 7%-19%higher than other methods,and rank optimization can improve the recognition rate by about 10%-17%.(2)This paper presents a gait recognition method based on Gauss mapping.Gait is a common biological feature of person re-identification.Its main advantage is that it can recognize pedestrians at a distance of 10 meters beyond the failure of other person re-identification technologies.The feature of the gait recognition method proposed in this paper is the combination of distance transformation and local contour curvature,which combines the internal contour and boundary curvature of the body into a new feature descriptor,which is more robust than the existing gait feature descriptor.The proposed method is evaluated on widely used and challenging USF,CASIA-B and OULP datasets.The results on USF dataset show that the average recognition rate of Rank-1 and Rank-5 are 54.8%and 71.3%respectively,which are higher than those of the comparative methods.The results on CASIA-B dataset show that the average recognition rate of the proposed method is the best in all different perspectives.The results on OULP gait dataset show that the average recognition rate of the proposed method is the best in four different perspectives,and can surpass the comparative method.(3)In order to achieve the practical application in the printing intelligent factory,a person re-identification system model is established.Firstly,the image in training set and test set is preprocessed,then the features are extracted from the pedestrian image,and the pedestrian feature representation model is established according to the acquired features.Finally,distance measurement is used to measure the distance,and the matching results are obtained.