High-Accuracy Traffic Signal and Road Sign Detection Using YOLOv11x on CPU-Based Platforms

Accurate and reliable detection of traffic signals and road signs is a fundamental requirement for intelligent transportation systems and advanced driver assistance applications.Although recent deep learning-based object detection models have achieved remarkable performance, most existing studies rely heavily on high-end GPU hardware, limiting their accessibility and real-world deployment on resource-constrained platforms.This paper presents a high-accuracy traffic signal and road sign detection framework based on the YOLOv11x architecture, developed and evaluated using modern CPU-based computing environments.The proposed approach leverages multi-scale feature enhancement and spatial attention mechanisms within YOLOv11x to improve the detection of small, distant, and partially occluded traffic objects.A composite dataset was constructed by integrating images from the Chinese Traffic Sign Database (CTSD), BDD100K, and KITTI datasets, resulting in a total of 16,164 labeled images spanning 140 traffic sign and signal categories.All experiments were conducted on a CPU-only platform, demonstrating the feasibility of training and deploying high-capacity detection models without specialized GPU hardware.