This research investigates the deployment of a quantized convolutional neural network for traffic sign recognition on an embedded FPGA platform. A convolutional neural network was trained on a dataset conforming to the Vienna Convention on Road Signs and Signals to demonstrate the benefits of international standardization in improving classification performance. The trained model was quantized using Brevitas to reduce precision, exported through QONNX, and compiled with the FINN framework for deployment on a PYNQZ2 board. A live webcam was integrated to simulate real-time image acquisition for inference. The deployed system achieved an accuracy of 94.45%, demonstrating the feasibility of low bit-width neural networks for real-time, low-latency inference on resourceconstrained hardware. This work highlights the critical role of quantization-aware training, model streamlining, and hardwaresoftware co-design in enabling efficient edge AI deployment.SDG alignment—This research advances SDG 3 (Good Health and Well-Being), target 3.6 and SDG 11 (Sustainable Cities and Communities),
target 11.2 by enabling robust driver-fatigue detection to improve road safety and reduce traffic injuries within safer, more sustainable transport systems.

FPGA, Convolutional Neural Network, Quantization, Traffic Sign Recognition, Edge AI, PYNQ, FINN