摘要
针对单一任务模型不能同时满足自动驾驶多样化感知任务的问题,提出了一种基于改进YOLOv5s的快速端到端道路多任务感知方法。首先,在YOLOv5s网络输出端设计两个语义分割解码器,能够同时完成交通目标检测、车道线和可行驶区域检测任务。其次,引入Rep VGG block改进YOLOv5s算法中的C3结构,借助结构重参数化策略提升模型速度和精度。为了提升网络对于小目标的检测能力,引入位置注意力机制对编码器的特征融合网络进行改进;最后基于大型公开道路场景数据集BDD100K进行实验验证该算法在同类型算法的优越性。实验结果表明,算法车辆平均检测精度为78.3%,车道线交并比为27.2%,可行驶区域平均交并比为92.3%,检测速度为8.03FPS,与同类型算法YOLOP、Hybrid Nets对比,该算法综合性能最佳。
Aiming at the problem that a single task model can not satisfy the diverse perception tasks of autonomous driving at the same time,a fast end-to-end road multi-task perception method based on improved YOLOv5s is proposed.First,two semantic segmentation decoders are designed at the output of the YOLOv5s network,which can simultaneously complete the tasks of traffic object detection,lane line and drivable area detection.Secondly,the RepVGG block is introduced to improve the C3 structure in the YOLOv5s algorithm,and the speed and accuracy of the model are improved with the help of the structural re-parameterization strategy.Then,in order to improve the detection ability of the network for small targets,the position attention mechanism is introduced to improve the feature fusion network of the encoder.Finally,based on the large-scale public road scene dataset BDD100K,experiments are carried out to verify the superiority of the proposed algorithm in the same type of algorithm.The experimental results show that the average vehicle detection accuracy of the algorithm in this paper is 78.3%,the lane line intersection ratio is 27.2%,the average drivable area intersection ratio is 92.3%,and the detection speed is 8.03FPS.Compared with the same type of algorithms YOLOP and HybridNets,this paper the algorithm has the best overall performance.
作者
宫保国
陶兆胜
赵瑞
李庆萍
伍毅
吴浩
GONG Bao-guo;TAO Zhao-sheng;ZHAO Rui;LI Qing-ping;WU Yi;WU Hao(College of Mechnical Engineering,Anhui University of Technology,Anhui Maanshan 243032,China)
出处
《齐齐哈尔大学学报（自然科学版）》
2023年第3期19-29,共11页
Journal of Qiqihar University(Natural Science Edition)
基金
安徽省自然科学基金面上项目(2108085ME166)
安徽高校自然科学研究项目重点项目(KJ2021A0408)。
关键词
无人驾驶
目标检测
多任务网络
autonomous driving
object detection
multi-task networks
