基于多尺度特征融合的乘用车路面信息预警研究方法  

Research Method of Passenger Vehicle Pavement Information Warning Based on Multi-scale Feature Fusion

作　　者：李香凝 韩昊 LI Xiangning;HAN Hao(Changchun University of Finance and Economics,Changchun 130112;Faw Pentium Car Co.,Ltd.,Changchun 130012,China)

机构地区：[1]长春财经学院,吉林长春130112 [2]一汽奔腾轿车有限公司,吉林长春130012

出　　处：《汽车电器》2024年第12期40-43,共4页Auto Electric Parts

摘　　要：伴随着用户对于乘用车行驶过程中的安全性和驾驶舒适性要求越来越高,计算机视觉算法作为一种先进的智能汽车技术被广泛应用。面对复杂多变的驾驶路面情况,实时准确的路面信息预警能够极大程度地保障用户出行。文章提出一种基于多尺度特征融合的乘用车路面信息预警方法。该方法在主流检测框架YOLOv8s中加入BiFPN方法,能有效提升对小目标的检测精度。此外,引入GIoU作为边界框回归损失函数的创新改进,更好地处理检测框不相交或重叠较少的情况,进一步提升模型在复杂路面情况下的目标定位精度。由于现有相关数据集并不完备,需进行数据采集和特征提取工作,这些数据应涵盖各种路面工况,以进一步满足路面信息预警需求。通过试验结果分析,本方法在乘用车路面信息检测方面表现出较好性能,不仅能增强对小目标的检测精度,也能提升边界框定位的鲁棒性,为后续的汽车路面信息预警研究提供有效的数据和技术支持。With the increasing requirements for safety and driving comfort of passenger cars,computer vision algorithm has been widely used as an advanced intelligent vehicle technology.In the face of complex and changeable driving road conditions,real-time and accurate road information early warning can greatly protect users'travel.This paper presents a road information warning method for passenger vehicles based on multi-scale feature fusion.The BiFPN method is added into the mainstream detection framework YOLOv8s,which can effectively improve the detection accuracy of small targets.In addition,GIoU is introduced as an innovative improvement of the boundary box regression loss function,which can better deal with the cases where the detection boxes do not intersect or overlap less,and further improve the target location accuracy of the model in complex road conditions.Due to the incompleteness of existing relevant data sets,data collection and feature extraction should be carried out,and these data should cover various road conditions to further meet the demand for road information early warning.Through the analysis of experimental results,the proposed method shows good performance in the detection of passenger vehicle road surface information,which can not only enhance the detection accuracy of small targets,but also improve the robustness of boundary frame positioning,providing effective data and technical support for the subsequent research on vehicle road surface information early warning.

关 键 词：乘用车 多尺度特征融合 路面信息 

分 类 号：U463.6[机械工程—车辆工程]