Monitoramento Rodoviário Inteligente via Redes Privativas e Satélite LEO: Uma Abordagem com SLMs e Aprendizado de Máquina Federado na Borda em Arquitetura Distribuída

This paper proposes an intelligent road monitoring architecture integrating private 5G networks, LEO satellites as backhaul infrastructure, and Edge AI with Federated Learning, addressing coverage gaps across 6,600 kilometers of Brazilian federal highways. Replacing raw video streaming with locally processed SLM model gradients at (RSUs) reduces data traffic by up to 96%, enabling continuous monitoring even in areas with intermittent connectivity. The architecture was validated in the CPQD laboratory environment with Starlink backhaul and a 5G C2n Core, and complemented by a comparative orbital telemetry study between the SP340 RMC (19.8 satellites/sample) and the BR230 Transamazônica (12.1 satellites/sample). The results confirmed the feasibility of the proposal in asymmetric satellite coverage contexts, ensuring compliance with LGPD and guaranteeing data confidentiality and integrity.