Sistema inteligente para la automatización del rastreo vehicular por medio de múltiples cámaras

Abstract

CONDITION FOR PUBLICATION OF PROJECT. project implements an automated vehicle tracking system using Deep Learning and computer vision techniques. The system employs YOLOv8 for vehicle detection, ByteTrack for intra-camera tracking, and Vision Transformer (ViT) to generate robust embeddings enabling multi-camera ReIDentification. Multiple camera video sequences were processed, applying cosine similarity matching and re-ranking techniques to improve accuracy. Results demonstrated 94.7% Rank-1 accuracy and 77.96% mAP, validating the effectiveness of the proposed approach. The system provides a scalable solution for automated vehicle monitoring in complex urban environments. Keywords: Vehicle tracking, Computer vision, Deep Learning, ReIDentification, Multi-camera processing.