Desarrollo de un sistema de seguimiento de personas mediante re-identificación inteligente

Abstract

DOCUMENT IN THE PROCESS OF INTELLECTUAL PROPERTY. Ecuador is facing an exponential rise in crime. As a response, video surveillance systems have been implemented to record and playback footage. The limitations of these systems include manual analysis that can be time consuming. At ESPOL, several departments have installed surveillance systems, but CIDIS lacks an efficient method to track and monitor its employees. In response, a person detection and tracking system based on re-identification algorithms and deep learning techniques is proposed, aiming to enhance security and indoor space management whilst considering the possibility of extending the system throughout the whole institution to strengthen its security level. To develop this system, three individual solutions were integrated into one platform to detect, classify and track a person in video footage, primarily using Vision Transformers and focusing on user-friendliness and accuracy. As a result, an 80% accuracy rate was achieved in distinguishing unique persons, while maintaining a 95% tracking precision. The system meets the established standards, is user-friendly and addresses the primary challenges at CIDIS. The system has potential to be implemented throughout ESPOL to reinforce its security measures. Keywords: re-identification, security, video tracking, system integration, deep learning.