针对智能驾驶场景下路面中小型障碍物易发生漏检、小目标障碍物难检测、模型参数量大等问题，提出了改进YOLOv8n的障碍物目标检测算法。在主干网络中融入分布移位卷积（distribution shifting convolution, DSConv），将浮点运算替换为整数运算，减少了冗余计算量，通过量化和分布移位的方式模仿原始卷积层，维持了准确率；通过添加小目标检测层，更好地捕捉小目标的特征信息，适配小目标的尺度特征；结合SimAM无参数注意力机制，引入SPPF-SimAM模块，提高特征表示的质量与多样性，在不增加参数量的情况下实现了检测精度的提升；通过组合鬼影混洗卷积（ghost-shuffle convolution, GSConv）和VoV-GSCSP模块的方式轻量化颈部特征融合网络，降低了模型的参数量和计算量。实验结果表明，改进后模型的准确率、召回率、平均精度均值相较于原始模型分别提升了1.6%、8.0%、6.2%，参数量降低了6.7%，所提算法有效提升了智能驾驶场景下中小型障碍物的检测精度，并且在检测性能与模型轻量化之间达到较好的平衡。

Aiming at the problems such as small and medium-sized obstacles on the road are prone to miss detection, small target obstacles are difficult to detect, and the number of model parameters is large in smart driving scenarios, the obstacle target detection algorithm with improved YOLOv8n was proposed. Distribution shifting convolution (DSConv) was used in the backbone network to replace floating point operation with integer operation, reducing the amount of redundant computation, and maintaining the accuracy by imitating the original convolution layer by quantization and distribution shifting. By adding small target detection layer, the feature information of small target can be captured better and the scale characteristics of small target can be adapted. Combined with SimAM parameterless attention mechanism, SPPF-SimAM module was introduced to improve the quality and diversity of feature representation, and the detection accuracy was improved without increasing the number of parameters. By combining ghost-shuffle convolution(GSConv) and VoV-GSCSP modules, the neck feature fusion network was lightweight, reducing the number of parameters and calculation of the model. The experimental results show that the accuracy, recall, and mean average precision of the improved model are improved by 1.6%, 8.0%, and 6.2%, respectively. The number of parameters is reduced by 6.7% compared with the original model, and the proposed algorithm effectively improves the detection accuracy of small and medium-sized obstacles in smart driving scenarios, and achieves a better balance between the detection performance and the model lightweighting.