Pedestrian Intelligent Recognition And Tracking System Under Cross-platform

In recent year,public safety has become an increasingly concerned issue in the area of social management.Since human identification is usually the key to public safety,it is very important to extract and analyze the effective pedestrian information from video surveillance network.However,due to the huge differences in the background,illumination and resolution of the image captured by different monitoring devices,it is difficult to effectively identify and track the target pedestrian in the video surveillance network.Therefore,it is very meaningful to design a pedestrian intelligent recognition and tracking system under cross-platform.In this paper,a face recognition technology,a pedestrian detection technology and a pedestrian identification and tracking technology are proposed for different application scenarios.The simulation results in the public databases show that these technologies proposed in this paper can achieve better results.The main works and innovations of this paper are as follows:In the first,for the face recognition problem,an improved subspace learning method is proposed.To be specific,a subspace learning model named truncated nuclear norm on low rank discriminant embedding(TNNL)is built to extract image low dimensional features,and sparse representation classifier is used to classify these low dimensional features,so as to realize face recognition.By adding truncated nuclear norm to better solve the rank minimization problem,TNNL can mitigate the negative impact of image noise.By imposing the linear discriminant term to transform the unsupervised model into the supervised model,TNNL can improve the discriminative ability of the low dimensional features.Simulation results show that TNNL can effectively extract the robust low dimensional features,and can obtain the high recognition rate.Next,for the pedestrian detection problem,an improved elastic network method is proposed.Because the traditional sparse descriptor has some disadvantages,such as redundant coding and insufficient representation of image features,which can impact the accuracy of pedestrian detection.The following improvements are made to solve the above problems.Firstly,a novel dictionary learning model which named low-rank dictionary learning(LRD)is built.It can learn a clean and compact pedestrian dictionary.Secondly,a new sparse representation model is built with name of sparse coding on similar constraints and elastic net(CES).It can eliminate useless columns in the dictionary which impact the accuracy of sparse coding.Thirdly,a novel sparse descriptor is constructed.Finally,by serial fusing the novel sparse descriptor with the histogram of oriented gradient(HOG)feature and the completed modeling of the local binary pattern(CLBP)feature,the new pedestrian descriptor named multi-feature fusion descriptor based on low-rank and constrained sparse representation(LRCSF)is obtained.LRCSF can extract different levels of information from the pedestrian image while retaining the main structure of the pedestrian.Simulation results show that LRCSF is highly efficacious for pedestrian detection.In the end,as the pedestrian identification and tracking problem,a pedestrian identification and tracking technology based on multi-feature fusion is proposed.Because the single feature cannot fully represent the image information,a new fusion feature named LOMO-VGG16 is constructed by serial fusing the local maximal occurrence(LOMO)feature and the depth feature with the VGG16 network.LOMO feature can effectively extract the color information and texture information in the image,while the depth feature can effectively extract the depth abstract information in the image.So,the LOMO-VGG16 feature can describe the image information more comprehensively.In the pedestrian identification,LOMO-VGG16 feature of the pedestrian target is extracted,and the pedestrian identity information is obtained by using the cross-view quadratic discriminant analysis(XQDA)metric learning method.In the pedestrian tracking,a particle filter tracking method based on LOMO-VGG16 feature is proposed to track the target pedestrian.Simulation results show that the constructed LOMOVGG16 feature in this paper is highly efficacious for pedestrian identification and tracking.