Research on Road Target Detection Method Based on YOLOv8n
佳宠 张 · Computer Science and Application · 2025
自动驾驶中的感知系统主要使用目标检测算法来获取障道路上目标的分布，以便进行识别和分析。当前的目标检测算法发展迅速，但在实际应用场景中平衡实时检测和高检测精度的要求具有挑战性。为了解决上述问题，本文使用YOLOv8n作为原始模型，并提出了一个名为YOLOv8n-CSS的目标检测网络。首先，引入CBAM混合注意力机制增强关键特征提取并去除冗余，从而提高网络对物体和背景的识别能力。然后，使用SPPCSPC模块替换原始模型骨干网络中的SPPF模块，能够更好地融合来自不同层次和尺度的特征信息，可以有效地捕捉不同尺度物体的特征，提高模型识别物体的准确性。最后，引入SPD-Conv模块替换原始的交错卷积层，进行下采样操作，保留了更多的特征信息，从而提高了不同尺度目标的检测能力。在KITTI数据集和BDD100K数据集上的实验结果表明，改进的网络模型的平均准确率分别达到96.1%和48.0%，比基线模型分别高出3.9%和7.9%，明显优于基线模型。该模型在保证高检测精度的基础上，可以实现一般场景中的实时图像处理。The perception system in autonomous driving mainly uses object detection algorithms to obtain the distribution of objects on the road for identification and analysis. Although current object detection algorithms are developing rapidly, it remains challenging to balance the requirements of real-time detection and high detection accuracy in practical application scenarios. To address the above issues, this paper uses YOLOv8n as the original model and proposes an object detection network named YOLOv8n-CSS. First the CBAM hybrid attention mechanism is introduced to enhance the extraction of key features and remove redundancy, thereby improving the network’s ability to distinguish objects from the background. Then, the SPPF module in the backbone network of the original model is replaced with the SPPCSPC module. This allows for better integration of feature information from different levels and scales, effectively capturing the features of objects of various scales and improving the accuracy of object recognition by the model. Finally, the SPD-Conv module is introduced to replace the original staggered convolution layer for downsampling operations, which retains more feature information and thus enhances the detection ability for objects of different scales. Experimental results on the KITTI dataset and the BDD100K dataset show that the average accuracy of the improved network model reaches 96.1% and 48.0% respectively, which is 3.9% and 7.9% higher than that of the baseline model, significantly outperforming the baseline model. This model can achieve real-time image processing in general scenarios while ensuring high detection accuracy.