基于PointPillars的多模态融合三维目标检测方法  

Multi-modal Fusion 3D Object Detection Method Based on PointPillars

作　　者：王睿智 张虹 WANG Ruizhi;ZHANG Hong(Taiyuan Normal University,Jinzhong 030619,China)

机构地区：[1]太原师范学院,山西晋中030619

出　　处：《现代信息科技》2025年第5期56-61,共6页Modern Information Technology

基　　金：太原师范学院研究生教育创新项目(SYYJSYC-2395)。

摘　　要：点云数据的稀疏性和无序性会导致相关检测算法在远小目标检测中容易出现漏检和误检的问题。因此,文章基于PointPillars算法提出了一种多模态融合的三维目标检测算法。该算法设计了一种多模态特征融合柱体化编码模块,能够融合点云特征和图像特征,从而增强特征的语义信息,提高远小目标的检测精度。在KITTI数据集上的实验结果显示,与基线模型相比,汽车、行人和骑行者类别的三维平均检测精度分别提高了3.28%、2.88%和1.62%。结果表明,所提出的基于PointPillars的多模态融合三维目标检测方法能够有效减少远小目标的误检和漏检。The sparsity and disorder of point cloud data lead to the problems of missed detection and false detection in the detection of distant and small objects.Therefore,this paper proposes a multi-modal fusion 3D object detection algorithm based on the PointPillars algorithm.The algorithm designs a multi-modal feature fusion pillar encoding module,which can fuse point cloud features and image features,thereby enhancing the semantic information of features and improving the detection accuracy of distant and small objects.The experimental results on the KITTI dataset show that compared with the baseline model,the 3D average detection accuracy of the vehicle,pedestrian and cyclist categories is improved by 3.28%,2.88%and 1.62%,respectively.The results show that the proposed multi-modal fusion 3D object detection method based on PointPillars can effectively reduce the false detection and missed detection of distant and small objects.

关 键 词：多模态融合 三维目标检测 柱体化编码 PointPillars 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391[自动化与计算机技术—控制科学与工程]