Person Localization using Machine Learning in Multi-Source Camera Surveillance System

The localization of individuals, using face recognition technology, is important in today’s society for various reasons, such as locating criminals. Closed-circuit television (CCTV) systems offer high level of security at a reasonable price by recognizing suspicious activities. Because of the wide usage of facial recognition technology, the general public’s worry about privacy increased significantly. These concerns have prompted numerous federal authorities to propose new legislation to protect personal privacy. In this paper, we propose an accurate person localization scheme to enable law enforcement agencies to identify the locations visited by wanted and suspected people using surveillance cameras placed in various public places. Unlike the existing techniques that measure the Euclidean distance between two images to determine if they are for the same person, we propose a more accurate technique that is based on a machine learning model to determine whether the features of two images belong to the same person. Our performance evaluations demonstrate that our model outperforms the Euclidean distance-based schemes when operating in a multi-source camera environment with changing recording quality.