Experimental Telemetry, Codebase, and 4K Vision Benchmarks for Edge-NPU Traffic Control

This repository contains the complete experimental dataset, open-source codebase, model compilation scripts, simulation telemetry logs, and hardware deployment drivers for the NPU-accelerated edge traffic management system associated with the paper "Edge-AI Embedded Real-Time Signal Control: Hardware-in-the-Loop Implementation of Webster-Driven NEMA TS-2 Logic on NPU Acceleration Architecture". The repository is organized into four modular components: 01_Microsimulation_Framework: 2D Kinematic IDM traffic microsimulator, 630-cycle non-stationary stress protocol generator, and 114,936 telemetry log records across 18 demand phases. 02_Offline_Verification: Numerical validation scripts for 282 mathematical boundary conditions of the Webster timing core, automated loss convergence analysis, and statistical significance tests (paired t-tests, Wilcoxon signed-rank, Cohen's d, Friedman test). 03_On_Device_Evaluation: NeuralONE SDK (Q4 release) compilation toolchain for converting raw ONNX YOLO26 detection models to dual-output INT8 .cix format using 64 real-world calibration frames, on-device NPU inference scripts, and cross-platform latency benchmarks across model scales (n, s, m, l, x). 04_Edge_Controller: Multi-threaded asynchronous edge orchestrator, 18-state NEMA TS-2 compliant Finite State Machine (FSM), Linux gpiod 16-channel LED relay hardware drivers, PyQt5 real-time telemetry dashboard GUI, and persistent JSON situational memory store.