Design of an Embedded Lightweight System for Traffic Signal Detection in Adverse Weather and at Night Based on YOLOXNano

Abstract

With the rapid advancement of intelligent transportation systems (ITS) and autonomous driving technologies, the robustness of traffic signal detection under complex meteorological and nighttime conditions has emerged as a critical concern. To address challenges such as diminished recognition accuracy in adverse weather and low-light conditions, as well as deployment constraints on edge devices, this paper proposes an embedded lightweight traffic signal detection system based on YOLOX-Nano. The system enhances environmental adaptability via a diversified meteorological scene data augmentation strategy, while achieving model lightweighting through network pruning and mixed-precision quantization, effectively balancing detection accuracy, real-time performance, and resource efficiency within the constraints of limited hardware resources [1]. During the deployment phase, the system utilizes ONNX and TensorRT to establish an efficient inference toolchain, enabling cross-framework model conversion and acceleration optimization, ultimately achieving low-power, high-performance real-time traffic signal detection on the Jetson Nano platform. The primary focus of this paper is on the system architecture design and its implementation pathway, aiming to provide an embedded lightweight solution with engineering feasibility and scalability for traffic signal detection under complex meteorological conditions.