基于点云掩码预训练与高斯定位不确定性估计的激光雷达目标检测方法

冯宇; 谢光达; 刘龙; 苏云泉; 刘珺玮; 耿艳东; 苗蕾

doi:10.12382/bgxb.2024.0788

您当前的位置：

首页 >

文章列表页 >

基于点云掩码预训练与高斯定位不确定性估计的激光雷达目标检测方法

更新时间：2025-09-05

- 基于点云掩码预训练与高斯定位不确定性估计的激光雷达目标检测方法
- LiDAR Object Detection Method Based on Point Cloud Mask Pre-training and Gaussian Localization Uncertainty Estimation
- 兵工学报 2025年46卷第6期页码：240788
- 作者机构：
  
  内蒙古第一机械集团股份有限公司科研所, 内蒙古包头 014000
- 作者简介：
  
  *邮箱:xgd_hhh@163.com
- 基金信息：
- DOI：10.12382/bgxb.2024.0788
  中图分类号：
- 收稿：2024-09-03，
  
  网络出版：2025-06-28，
  
  纸质出版：2025-06-10
- 稿件说明：
移动端阅览
冯宇, 谢光达, 刘龙, 等. 基于点云掩码预训练与高斯定位不确定性估计的激光雷达目标检测方法[J]. 兵工学报, 2025,46(6):240788.

Yu FENG, Guangda XIE, Long LIU, et al. LiDAR Object Detection Method Based on Point Cloud Mask Pre-training and Gaussian Localization Uncertainty Estimation[J]. Acta Armamentarii, 2025, 46(6): 240788.
冯宇, 谢光达, 刘龙, 等. 基于点云掩码预训练与高斯定位不确定性估计的激光雷达目标检测方法[J]. 兵工学报, 2025,46(6):240788. DOI： 10.12382/bgxb.2024.0788.

Yu FENG, Guangda XIE, Long LIU, et al. LiDAR Object Detection Method Based on Point Cloud Mask Pre-training and Gaussian Localization Uncertainty Estimation[J]. Acta Armamentarii, 2025, 46(6): 240788. DOI： 10.12382/bgxb.2024.0788.

摘要

激光雷达获取的三维点云数据是自动驾驶的关键数据源

但其标注难度大、数据量有限且标签不确定性高

限制了基于深度学习的三维目标检测模型的训练效果。针对以上问题

提出点云掩码策略构建预训练数据集

结合迁移学习提升模型检测精度;提出基于高斯分布的定位不确定性估计建模方法

使目标检测模型在预测边界框坐标的同时预测每个坐标的定位不确定性。实验结果表明

在不明显增加算法复杂度的前提下

新方法有效减少了误检现象

显著提高了目标检测的准确性。

Abstract

The 3D point cloud data acquired by LiDAR is crucial for autonomous driving.However

the annotation of point cloud data is difficult;the available data is limited;and there is usually a high uncertainty in its labels

which constrain the training effects of deep learning-based 3D object detection models.To address these issues

this paper proposes a point cloud masking strategy to construct a pre-training dataset

which is combined with transfer learning to improve detection accuracy.Additionally

a Gaussian distribution-based localization uncertainty estimation modeling method is proposed

enabling the object detection model to predict the localization uncertainty of each coordinate while predicting the bounding box coordinates.Experimental results demonstrate that the proposed method effectively reduces the false detections and significantly improves the accuracy of object detection without significantly increasing algorithmic complexity.

关键词

Keywords

references

龚诗雄 , 王旭 , 孔国杰 , 等 . 多车协同目标跟踪方法 [J ] . 兵工学报 , 2022 , 43 ( 10 ): 2429 - 2442 .

GONG S X , WANG X , KONG G J , et al . Multi-vehicle cooperative target tracking method [J ] . Acta Armamentarii , 2022 , 43 ( 10 ): 2429 - 2442 . (in Chinese)

刘桂宇 . 面向激光雷达点云的三维目标识别算法研究综述 [J ] . 信息记录材料 , 2022 , 23 ( 4 ): 243 - 245 .

LIU G Y . A review of 3D object recognition algorithms for LiDAR point clouds [J ] . Information Recording Materials , 2022 , 23 ( 4 ): 243 - 245 . (in Chinese)

霍威乐 , 荆涛 , 任爽 . 面向自动驾驶的三维目标检测综述 [J ] . 计算机科学 , 2023 , 50 ( 7 ): 107 - 118 . DOI: 10.11896/jsjkx.220700090 http://doi.org/10.11896/jsjkx.220700090 近年来,随着自动驾驶行业的蓬勃发展,作为感知系统核心的三维目标检测技术受到越来越多的关注,已成为当前热门的研究方向。同时,深度学习的广泛应用,使得最近的三维目标检测技术有了很大的突破,大批优秀的算法涌现。文中系统地总结了面向自动驾驶领域的三维目标检测方法,并按传感器类型将现有的算法分为3类,即基于图像的三维目标检测、基于LiDAR的三维目标检测和基于多传感器的三维目标检测;其次,详细分析了3种方法的优缺点,并对基于LiDAR的三维目标检测算法进行了深入调研和细分;然后,介绍了自动驾驶领域常用的三维目标检测数据集,包括KITTI,nuScenes和Waymo Open Dataset,并对比了最新的三维目标检测算法在不同数据集上的性能表现;最后探讨了三维目标检测技术未来的发展方向。

HUO W L , JING T , REN S . A review of 3D object detection for autonomous driving [J ] . Computer Science , 2023 , 50 ( 7 ): 107 - 118 . (in Chinese) DOI: 10.11896/jsjkx.220700090 http://doi.org/10.11896/jsjkx.220700090 In recent years,with the rapid development of autonomous driving,3D object detection technology as the core of perception systems has received more and more attention and become a hot research direction.At the same time,the wide application of deep learning has made a great breakthrough in 3D object detection technology recently.A large number of excellent algorithms have emerged.This paper systematically summarizes 3D object detection methods for the autonomous driving field and divides the existing algorithms into three categories according to sensor types:image-based 3D object detection,LiDAR-based 3D object detection,and multi-sensor-based 3D object detection.After that,it analyzes the advantages and disadvantages of the three methods in detail.The LiDAR-based 3D object detection algorithms are thoroughly investigated and subdivided.Then it introduces the commonly used 3D object detection datasets in autonomous driving,including KITTI,nuScenes,and Waymo Open Dataset,and compares the performance of the latest 3D object detection algorithms on different datasets.Finally,the future research direction of 3D object detection technology is discussed.

VASCONCELOS C , BIRODKAR V , DUMOULIN V . Proper reuse of image classification features improves object detection [C ] // Proceedings of Conference on Computer Vision and Pattern Recognition.New Orleans,LA,US:IEEE , 2022 : 13628 - 13637 .

ZHANG R R , GUO Z Y , FANG R Y , et al . Point-M2AE:multi-scale masked autoencoders for hierarchical point cloud pre-training [J ] . Advances in Neural Information Processing Systems , 2022 , 35 : 27061 - 27074 .

CAESAR H , BANKITI V , LANG A H , et al . nuscenes:a multimodal dataset for autonomous driving [C ] // Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Seattle,WA,US:IEEE , 2020 : 11621 - 11631 .

汪萌 , 诸兵 . 不确定性建模在2D和3D目标检测中的应用 [J ] . 系统工程与电子技术 , 2023 , 45 ( 8 ) : 2370 - 2376 . DOI: 10.12305/j.issn.1001-506X.2023.08.10 http://doi.org/10.12305/j.issn.1001-506X.2023.08.10 目标检测算法在自动驾驶领域有着不可或缺的地位, 其检测精度和速度往往可以作为评判一个自动驾驶系统好坏的标准。如何提升目标检测精度和速度已成为当前目标检测算法的主要研究方向。对此, 提出了一种基于不确定性建模的目标检测改进算法, 在原有二维单次多柜检测器上通过对物体的边界框进行高斯建模并引入新的位置损失函数, 实现对原有检测结果的微调。同时, 在原有单阶单目三维检测器的基础上引入深度以及航向角的不确定性来调整目标在热力图中的高斯半径, 并用马氏距离替代原有的L1距离以提升对较远目标和斜向目标的识别率。对比实验证明, 改进后的2D和3D检测算法在KITTI自动驾驶数据集上对比原始2D和3D算法, 检测精度分别提升了近5%和2%。

WANG M , ZHU B . Application of uncertainty modeling in 2D and D target detection [J ] . Systems Engineering and Electronics , 2023 , 45 ( 8 ) : 2370 - 2376 . (in Chinese)

XU X W , WANG Y F , ZHENG Y , et al . Back to reality:weakly-supervised 3D object detection with shape-guided label enhancement [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.New Orleans,LA,US:IEEE , 2022 : 8438 - 8447 .

MEYER G , THAKURDESA N . Learning an uncertaintyaware object detector for autonomous driving [C ] // Proceedings of IEEE International Conference on Intelligent Robots and Systems.Las Vegas,NV,US:IEEE , 2020 : 10521 - 10527 .

HE K M , CHEN X L , XIE S N , et al . Masked autoencoders are scalable vision learners [C ] // Proceedings of Conference on Computer Vision and Pattern Recognition.New Orleans,LA,US:IEEE , 2022 : 16000 - 16009 .

GEIGER A , LENZ P , STILLER C , et al . Vision meets robotics:the KITTI dataset [J ] . International Journal of Robotics Research , 2013 , 32 ( 11 ): 1231 - 1237 .

SUN P , KRETZSCHMAR H , DOTIWALLA X , et al . Scalability in perception for autonomous driving:Waymo open dataset [C ] // Proceedings of IEEE Conference on Computer Vision and Pattern Recognition.Seattle,WA,US:IEEE , 2020 : 2446 - 2454 .

CHEN Y , LIU J , ZHANG X , et al . Voxelnext:fully sparse voxelnet for 3d object detection and tracking [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.Vancouver,BC,Canada:IEEE , 2023 : 21674 - 21683 .

HU R , CAO F L , WANG W J . A novel complementary dual-aware network for point cloud classification [J ] . Engineering Applications of Artificial Intelligence , 2024 , 137 : 109224 .

YAN Y , MAO Y X , LI B . Second:Sparsely embedded convolutional detection [J ] . Sensors , 2018 , 18 ( 10 ): 3337 - 3354 .

谢兄 , 杨金鹏 . YOLO-wLU:考虑定位不确定性的目标检测算法 [J ] . 计算机工程与应用 , 2021 , 57 ( 22 ): 223 - 231 . DOI: 10.3778/j.issn.1002-8331.2007-0234 http://doi.org/10.3778/j.issn.1002-8331.2007-0234 YOLOv3目标检测算法在检测目标时没有考虑边界框坐标定位存在的不确定性，因此有时不能得出正确的检测结果。针对此问题，提出YOLO-wLU（YOLO with Localization Uncertainty）算法。该算法借鉴深度学习中的不确定性思想，使用高斯分布函数建立边界框坐标的概率分布模型以考虑边界框坐标定位不确定性；设计新的边界框损失函数，在检测过程中移除定位不确定性较大的检测结果；通过融合周围边界框坐标信息提高了边界框坐标辨识结果的准确性。实验结果表明，该算法可有效减少误报率，提高检测精度；COCO数据集上测试结果显示，相比YOLOv3算法，该算法的mAP最高可提升4.1个百分点。

XIE X , YANG J P . YOLO-wLU:an object detection algorithm considering localization uncertainty [J ] . Computer Engineering and Applications , 2021 , 57 ( 22 ): 223 - 231 . (in Chinese)

PSAROS A F , MENG X H , ZOU Z R , et al . Uncertainty quantification in scientific machine learning:Methods,metrics,and comparisons [J ] . Journal of Computational Physics , 2023 , 477 : 111902 .

WANG Z , DAHL G E , SWERSKY K , et al . Pre-trained Gaussian processes for Bayesian optimization [J ] . Journal of Machine Learning Research , 2024 , 25 ( 212 ): 1 - 83 .

LIANG A , HUA H Y , FANG J , et al . Boosting 3D point-based object detection by reducing information loss caused by discontinuous receptive fields [J ] . International Journal of Applied Earth Observation and Geoinformation , 2024 , 132 : 104049 .

KONG X H , LI X J , ZHU X X , et al . Detection model based on improved faster-RCNN in apple orchard environment [J ] . Intelligent Systems with Applications , 2024 , 21 : 200325 .

KINGMA D , BA J . Adam:a method for stochastic optimization [C ] // Proceedings of the 3rd International Conference on Learning Representations.San Diego,CA,US:ICLR , 2015 : 1 - 15 .

MARK E , LUC V , CHRISTOPHER K , et al . The PASCAL visual object classes (VOC) challenge [J ] . International Journal of Computer Vision , 2010 , 88 : 303 - 338 .

YIN T W , ZHOU X Y , KRAHENBUHL P . Center-based 3D object detection and tracking [C ] // Proceedings of IEEE International Conference on Computer Vision.Nashville,TN,US:IEEE , 2021 : 11784 - 11793 .

CHEN Y K , LI Y W , ZHANG X Y , et al . Focal sparse convolutional networks for 3D object detection [C ] // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.New Orleans,LA,US:IEEE , 2022 : 5428 - 5437 .

浏览量

下载量

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

非结构化环境下适用多类型激光雷达的地面无人车辆负障碍物检测

基于动态区域剔除与稠密地图构建的视觉SLAM算法

基于图神经网络的车辆目标遮蔽关重部位检测

基于红外相机和毫米波雷达融合的烟雾遮挡无人驾驶车辆目标检测与跟踪

基于MSMA-YOLO的破片群目标检测