Study on Person Re-Identification Methods in Public Surveillance Systems

Person re-identification (re-id) means it's a process of identifying a person which appears once in a camera of any public places. That means it aims to find a person captured in one camera in another camera at any new location. Person re-identification has many important security related applications such as cameras in shopping malls and stores. One of the major challenges in person re-identification method is the lack of spatial and temporal changes. The main aim is to compare different methods of re-id which includes an end-to- end Accumulative Motion Context Network (AMOC) based method addressing video person re- identification problem through joint spatial appearance learning , an image-to- video person re-id which is different from the most existing research of person re-id in probe and gallery setting, a visual attention model that is formulated as a triplet recurrent neural network which takes several glimpses of triplet images of persons and dynamically generates comparative attention location maps for person re-identification. The last two methods are about deep Convolutional Neural Networks (PersonNet) and reference descriptor which uses pre-defined set of values for re-id. The study incorporates each and every method in detail along with a comparative analysis aimed to realize advantages and disadvantages of above mentioned approaches.