Abstract: In view of the lack of comfort for vehicle autonomous driving in obstacle target detection, and the problem that speed bumps, as a common traffic facility in daily life, will affect vehicle comfort, this paper realizes speed bump detection under good light conditions based on YOLOv8n improved algorithm. First, the speed bump data set is constructed at different distances and in different environments under good lighting conditions. Secondly, the SimAM attention mechanism is added on the basis of YOLOv8n to improve the network's perception of feature maps. Finally, the original loss function CIoU is replaced by SIoU to improve the convergence speed and robustness of the network. The results show that the mean average accuracy of the improved model mAP@0.5 is increased by 4.3%. The effectiveness of the improved model is successfully verified through the detection of deceleration belts by experimental vehicles at different distances, and the detection scheme can accurately provide the pre-detection information of deceleration belts.

Key words: YOLOv8n; speed bump detection; SimAM; SIoU

摘要： 针对汽车自动驾驶在障碍物目标检测缺少对汽车自动驾驶舒适性，且减速带作为日常 生活中常见的交通设施会影响到车辆的舒适性的问题，文章基于 YOLOv8n 改进算法实现在 良好光线条件下减速带检测。首先，构建在光线良好条件下不同距离和不同环境下的减速带 数据集；其次，在 YOLOv8n 的基础上添加 SimAM 注意力机制，提高网络对特征图的感知； 最后，将原网络损失函数 CIoU 替换为 SIoU，提高网络收敛速度及鲁棒性。结果表明，改进 后的模型的均值平均精度 mAP@0.5 提高 4.3%，通过实验小车在不同距离下对减速带的检测 进行验证，成功验证了改进模型的有效性，检测方案能准确提供减速带检测的预前信息。

关键词: YOLOv8n；减速带检测；SimAM；SIoU