International Journal of Engineering Applied Sciences and Technology, 2021 Vol. 6, Issue 4, ISSN No. 2455-2143, Pages 361-364 Published Online August 2021 in IJEAST (http://www.ijeast.com) 361 PERSON RE-IDENTIFICATION USING DEEP METRIC LEARNING Ms. Shruti Jalapur Ms.Bibi Ayesha Hundekar Department of CSE Department of CSE SECAB Institute Vijayapura, Karnataka, India SECAB Institute Vijayapura,Karnataka,India Abstract— Today everyone faces a multitude and variety of threats ranging from robbery, kidnapping and terrorism to murder. To avoid these threats, authorities need to collect real-time information about what's going on in and around the city. New technologies are therefore being developed to make cities safer and more risk-free. Here we have built a reliable system that recognizes the person from every angle from a recorded image. We can get the input into the systems through CCTV cameras installed in public places where these types of life threatening events take place. It is easy to install these cameras in public places, and it is easier to monitor and store the data. The developed system uses deep metric learning and the machine learning platform, Tensor Flow and Keras. It's a type of machine learning where the system iteratively performs calculations to know the patterns. The system processes recorded images and compares them with existing data records in order to identify the person. The comparison is made based on certain selected features. The results are more accurate (98.18%) compared to existing systems. Keywords— Person Re-identification, Deep metric Learning, CNN I. INTRODUCTION An Introduction to Machine Learning: Machine learning is a branch of Artificial Intelligence (AS). For this reason, machine learning focuses on developing computer programs that can access and use data to learn on their own. Machine learning has been a buzzword for several years Arthur Samuel, an American IB Mer and pioneer in the field of computer games and artificial intelligence, coined the phrase machine learning in 1959. In 1997, Tom M. Mitchell published a widely referenced, more formal explanation of the machine learning algorithms: “In regard to a task T, a computer programme should learn from experience E. and quantify some performance. Machine learning is a type of deep metric learning. Deep metric learning (DML) is a distance-metric learning paradigm that combines deep learning and metric learning.. this requires that computer-systems perform iteratively calculate to identify patterns themselves. This means after a deep learning computer recognize that an image it is evaluated has the shape of a rectangle, oval or pattern by going through n cycles, and therefore our project will use the same logic as that Person re-identification applied via deep metric learning, in which it goes through n cycles to identify a person by assessing the similarities in relation to the data provided and finally presenting the positive results. To make this possible, we use Torchried, the pytorch-based library for deep metric learning, which is specifically considered for the identification of a person. II. PROPOSED ALGORITHM Implementation of this project is divided into three modules 1. Dataset preparation:  Three classes are made namely a,b and c  The class ‘a’ contains images of x person, class ‘b’ contains images of y person and class c contains images of ‘z’ person  All the classes are trained with their respective images around a data of 1400 images are trained  And a own dataset is created 2. Training the images:  After collecting the database the next procedural step is to tain thedata  This is an major act to gather the data by collecting the images and training atleast for 1 hour for a good accuracy  Duplicating the images hence for the better learning of the machine 3. Analyzing the data:  In this step we analyse the data and try to match it out with the present database  And once the needed picture matches with the given database the final upshot can be presented  And hence the last step is the final decision of identifying a person We used the transfer-learning technique, Mobilenet with image net weights as the base model with frozen upper layers, we added new layers to the base model and refined them on our user-defined data set, 1. MobileNet is an image classification model, it is a neural network with dense convolutional layers and pooling / subsampling layers. Compared to other