基于改进单步多框目标检测的道路小目标检测算法

针对SSD目标检测算法运用于自动驾驶领域时，在检测道路上小目标容易发生漏检错检的情况，本文提出一种改进的SSD目标检测算法。本算法首先在SSD模型的主干网络中嵌入感受野增强模块，扩大特征层的感受野，以获取更多小目标的特征信息；然后在主干网络后加入4次U型特征提取结构，构建4个不同层级的特征金字塔，最后合并成一个多层级特征金字塔用于检测。结果表明，该改进SSD模型在KITTI数据集上的检测精度较原始SSD模型提升了6%，检测速度达到了每秒27.9帧。在兼顾检测效率的同时，有效提高了对道路上小目标的检测精度，更适用于自动驾驶领域。

Road Small Target Detection Algorithm Based on Improved SSD

Aiming at the situation that the SSD target detection algorithm is prone to miss detection and wrong detection when detecting small targets on the road when applied to the field of autonomous driving, an improved SSD target detection algorithm was proposed. Firstly, the receptive field enhancement module is embedded in the backbone network of the SSD model to expand the receptive field of the feature layer to obtain more feature information of small targets; then four U-shaped feature extraction structures are added after the backbone network to construct four feature pyramids of different levels, and finally merged into a multi-level feature pyramid for detection. The results show that the improved SSD model improves the detection accuracy by 6% on the KITTI dataset compared with the original SSD model, and the detection speed reaches 27.9 frames per second. While taking into account the detection efficiency, it effectively improves the detection accuracy of small targets on the road and is more applicable to the field of autonomous driving.