SinoXivHardware-Accelerated 4-Way Traffic Management: An Adaptive Learning Control System using NeuralONE-Optimized YOLO26 and 4K Vision on Orange Pi 6 Plus NPU
Abstract
This paper presents a proof-of-concept for an urban traffic management system utilizing the hardware-accelerated capabilities of the Orange Pi 6 Plus. The proposed architecture leverages the CIX CD8180 SoC, which provides a combined AI computing power of 45 TOPS (CPU+NPU+GPU), enabling simultaneous processing of four 4K video streams at the edge using NeuralONE-optimized YOLO26 models for real-time vehicle detection and density analysis. The system implements a two-layer adaptive control algorithm: a deterministic Webster engine for phase timing optimization and an online learning mechanism driven by a physics-informed cost function. Experimental validation follows a controlled Proof of Concept (PoC) methodology using pre-recorded 4K intersection footage and a purpose-built traffic simulator with Intelligent Driver Model (IDM) physics, a NEMA TS-2 compliant 18-state controller, and an AI optimizer. Results demonstrate that the platform achieves NPU inference latencies of approximately 15-67 ms per frame depending on model scale, with stable multi-stream operation. The adaptive algorithm achieved cost convergence (ΔJ < 0) in 11 of 18 stress-test phases and passed all 282 mathematical boundary conditions, offering a scalable solution for smart city infrastructure in resource-constrained environments.
Keywords
Citation Information
@article{rogergeanycristianripaarias2026,
title={Hardware-Accelerated 4-Way Traffic Management: An Adaptive Learning Control System using NeuralONE-Optimized YOLO26 and 4K Vision on Orange Pi 6 Plus NPU},
author={Roger Geany Cristian Ripa Arias and Gonzalo Rafael Carpio Ramos and Franco Alessandro Arenas Mamani and German Alberto Echaiz Espinoza},
journal={Iranian Journal of Science and Technology, Transactions of Electrical Engineering},
year={2026},
doi={https://doi.org/10.21203/rs.3.rs-9373342/v1}
}
