Autonomous driving, as part of the intelligent transportation system, is increasingly becoming prevalent in people’s daily lives, and the ability to correctly recognize traffic signs is a crucial step in autonomous driving technology. Traditional traffic sign detection techniques primarily rely on color and shape-based methods. However, due to the influence of complex real-world factors such as weather and visibility, the detection performance often falls short of expectations. This paper introduces a new detection model called YOLOv8_ODY, built upon YOLOv8. To address the challenge of detecting small-sized traffic signs, YOLOv8_ODY incorporates a novel approach in its model backbone by introducing the full-dimensional dynamic convolution, referred to as ODConv. ODConv learns complementary attention from four dimensions of the convolutional kernel space, significantly enhancing the model’s ability to capture traffic sign objects. Furthermore, to improve the detection performance of the model’s head, a dynamic detection framework called DyHead is introduced, enhancing the perception of spatial position, spatial size, and task regions within the head. Experimental results demonstrate that, without altering the model’s depth, YOLOv8_ODY achieves a 2.6% improvement in mAP@0.5 compared to the original model. On the TT100k dataset, YOLOv8_ODY achieves an mAP@0.5 of 87.7%.