Person Re-Identification in Multi-Camera Environment

Abstract

The task of re-identifying the same person in several video/image sequences obtained from multiple camera viewpoints and angles is known as person Re-Identification (Re-ID). Person re-identification in a multi-camera system remains a difficult task. Researchers have done a lot of work on people re-id models, but each has its own set of limitations and needs improvement. The complexity of the architecture used also plays a significant role in re-identification in real-time scenarios. We demonstrate using a CNN-based feature extraction framework to re-identify persons in multi-camera environments based on their appearance. We present an approach for performing Re-Idtasks using a less dense CNN-backed model that employs Cross-input Neighborhood Difference (CND). The Re-ID model is trained and tested experimentally on the CUHK03 dataset, which contains diverse data. In this study, we look at the Re-ID task as a binary comparing task that determines if a given pair of images is similar or not. In comparison to state-of-the-art approaches, we demonstrate our proposed method on the CUHK03 dataset and attain rank-1 accuracy of 86.6 percent.