高速公路上无人驾驶汽车目标检测算法的研究.pdf

第40卷第1期 算 机 仿 真 2023年1月 文章编号 9348(2023)01-0137-06 高速公路上无人驾驶汽车目标检测算法的研究 葛雯，马 乐，屈乐乐 （沈阳航空航天大学电子信息工程学院，辽宁沈阳110136） 摘要：为了解决高速公路环境下无人驾驶汽车对目标检测准确度及速度的问题，提出一种改进SSD模型的目标检测算法， 通过结合空洞卷积改进DenseNet网络模型，构成D-DenseNet网络模型，并以此代替传统SSD算法中的VGG16 网络，从而简 化模型，减少运算量，提高检测精度，实现对目标快速且准确的检测。利用相关道路交通图像并结合深度学习中常用的 VOC数据集作为样本进行训练和测试，并与FasterR-CNN、YOLOv3及原始SSD模型进行比较，实验结果表明，与其它算法 相比，该算法对于高速公路不同环境下无人驾驶汽车的目标检测速度更快，精度更高，具有良好的鲁棒性。 关键词：无人驾驶；目标检测；仿真 中图分类号：TP391 文献标识码：B Research on Object Detection Algorithm for Unmanned Vehicles on Freeway GE Wen,MA Le,QU Lele ( College of Electronic Information Engineering, Shenyang Aerospace University , Shenyang Liaoning 110136, China) ABSTRACT:In order to solve the problem of target detection accuracy and speed of unmanned cars in highway envi- ronments ,this thesis proposes an improved object detection algorithm for the SSD model. The DenseNet network model is improved by integrating Dilated-Convolution to make up a D-DenseNet network model, which is an alternative to the traditional SSD algorithm VGG16 network, so as to predigest the model, reduce the amount of calculations, enhance the detection accuracy ,and achieve the rapid and accurate detection. Using relevant road traffic images and combining VOC datasets commonly used in deep learning as samples for training and testing,and comparing with Fas- ter R-CNN, YOLOv3 and primitive SSD models,the experimental results display that the algorithm has faster speed, higher precision and good robustness for target detection of unmanned vehicles on the freeway under different circum- stances. KEYWORDS:Unmanned vehicles; Object detection; Simulation