DGA-YOLO:基于深度与几何感知的自动驾驶目标检测

doi:10.16088/j.issn.1001-6600.202606032501

摘要/Abstract

摘要： 针对自动驾驶中目标尺度多变、环境复杂导致的检测精度低和鲁棒性不足问题,本文提出一种基于YOLO26改进的深度与几何感知目标检测算法DGA-YOLO (depth and geometry aware YOLO)。首先,设计R-G-Depth早期融合与迭代深度补全策略强化三维空间感知;其次,在主干引入语义-几何聚合器(semantic-geometric aggregator,SGA),双流并行提取局部语义与宏观几何上下文;再次,构建各向异性几何卷积(anisotropic geometric convolution,AGC),并联深度可分离与逐点卷积提取各向异性特征,结合容量下界约束的尺度感知注意力;最后,在融合层部署大核选择重构器(large selective reconstructor,LSR),以大核空洞卷积平滑噪声、十字条形卷积提纯几何特征。实验表明,该算法在KITTI上的平均精度均值(mAP)中的mAP@50与mAP@50∶95较基准算法分别提升2.5与4.4个百分点,在物理模拟雨、雾、雨雾交加极端环境与零样本泛化实验中均显著优于基准算法,鲁棒性与泛化能力优异。

关键词: 自动驾驶, 目标检测, YOLO26, 多模态融合, 深度与几何感知

Abstract: To address the problems of low detection accuracy and insufficient robustness caused by variable target scales and complex environments in autonomous driving, an improved depth and geometry aware object detection algorithm based on YOLO26, DGA-YOLO (depth and geometry aware YOLO), is proposed. Firstly, an R-G-Depth early fusion and iterative depth completion strategy is designed to enhance 3D spatial perception. Secondly, a Semantic-Geometric Aggregator (SGA) is introduced into the backbone to extract local semantics and macroscopic geometric context in parallel through two streams. Thirdly, an Anisotropic Geometric Convolution (AGC) is constructed, which parallels depthwise separable and pointwise convolutions to extract anisotropic geometric features, combined with a capacity-bounded scale-aware attention. Finally, a Large Selective Reconstructor (LSR) is deployed in the fusion layer, smoothing noise with large-kernel dilated convolutions and purifying geometric features with cross-strip convolutions. Experiments show that the proposed algorithm improves the mean Average Precision (mAP) on the KITTI dataset, with mAP@50 and mAP@50∶95 exceeding the baseline by 2.5 and 4.4 percentage points, respectively; it also significantly outperforms the baseline under physically simulated rain, fog, and rain-fog extreme conditions and in zero-shot generalization tests, demonstrating excellent anti-interference robustness and generalization capability.

Key words: autonomous driving, object detection, YOLO26, multi-modal fusion, depth and geometry awareness

中图分类号: TP391.41

李桂伸, 彭杰梁, 敖银辉. DGA-YOLO:基于深度与几何感知的自动驾驶目标检测[J]. 广西师范大学学报（自然科学版）, 2026, 44(6): 27-42.

Li Guishen, Peng Jieliang, Ao Yinhui. DGA-YOLO: depth and geometry-aware object detection for autonomous driving[J]. Journal of Guangxi Normal University(Natural Science Edition), 2026, 44(6): 27-42.

参考文献

[1] 任柯燕, 谷美颖, 袁正谦, 等. 自动驾驶3D目标检测研究综述[J]. 控制与决策, 2023, 38(4): 865-889. DOI: 10.13195/j.kzyjc.2022.0618.
[2] 罗会兰, 陈鸿坤. 基于深度学习的目标检测研究综述[J]. 电子学报, 2020, 48(6): 1230-1239. DOI: 10.3969/j.issn.0372-2112.2020.06.026.
[3] Law H, Deng J. CornerNet: detecting objects as paired keypoints[C]//Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018: 765-781. DOI: 10.1007/978-3-030-01264-9_45.
[4] Duan K W, Bai S, Xie L X, et al. CenterNet: keypoint triplets for object detection[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Los Alamitos, CA: IEEE Computer Society, 2019: 6569-6578. DOI: 10.1109/ICCV.2019.00667.
[5] Liu W, Anguelov D, Erhan D, et al. SSD: single shot multibox detector[C]//Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 21-37. DOI: 10.1007/978-3-319-46448-0_2.
[6] Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2016: 779-788. DOI: 10.1109/CVPR.2016.91.
[7] Ren S Q, He K M, Girshick R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI: 10.1109/TPAMI.2016.2577031.
[8] 赵洋, 王潇, 蔡柠泽, 等. 自动驾驶目标检测不确定性估计方法综述[J]. 汽车工程学报, 2024, 14(5): 760-771. DOI: 10.3969/j.issn.2095-1469.2024.05.02.
[9] 杨锋, 丁之桐, 邢蒙蒙, 等. 深度学习的目标检测算法改进综述[J]. 计算机工程与应用, 2023, 59(11): 1-15. DOI: 10.3778/j.issn.1002-8331.2209-0312.
[10]Peng L, Xu J K, Cheng H R, et al. Learning occupancy for monocular 3D object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2024: 10281-10292. DOI: 10.1109/CVPR52733.2024.00979.
[11]Qian R, Garg D, Wang Y, et al. End-to-end pseudo-LiDAR for image-based 3D object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2020: 5880-5889. DOI: 10.1109/CVPR42600.2020.00592.
[12]Liu Y, Sun P, Wergeles N, et al. A survey and performance evaluation of deep learning methods for small object detection[J]. Expert Systems with Applications, 2021, 172: 114602. DOI: 10.1016/j.eswa.2021.114602.
[13]张新钰, 邹镇洪, 李志伟, 等. 面向自动驾驶目标检测的深度多模态融合技术[J]. 智能系统学报, 2020, 15(4): 758-771. DOI: 10.11992/tis.202002010.
[14]Chen X Z, Ma H M, Wan J, et al. Multi-view 3D object detection network for autonomous driving[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2017: 6526-6534. DOI: 10.1109/CVPR.2017.691.
[15]Ku J, Mozifian M, Lee J, et al. Joint 3D proposal generation and object detection from view aggregation[C]//2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway, NJ: IEEE, 2018: 1-8. DOI: 10.1109/IROS.2018.8594049.
[16]李珣, 张友兵, 周奎, 等. 基于BEV多模态特征融合的三维目标检测算法[J]. 计算机应用研究, 2025, 42(12): 3823-3831. DOI: 10.19734/j.issn.1001-3695.2025.03.0101.
[17]Liu H B, Wu C, Wang H J. Real time object detection using LiDAR and camera fusion for autonomous driving[J]. Scientific Reports, 2023, 13: 8056. DOI: 10.1038/s41598-023-35170-z.
[18]Zhao X M, Sun P P, Xu Z G, et al. Fusion of 3D LIDAR and camera data for object detection in autonomous vehicle applications[J]. IEEE Sensors Journal, 2020, 20(9): 4901-4913. DOI: 10.1109/JSEN.2020.2966034.
[19]Winter J, Nowak R. Point cloud densification algorithm for multiple cameras and lidars data fusion[J]. Sensors, 2024, 24(17): 5786. DOI: 10.3390/s24175786.
[20]张介嵩, 黄影平, 张瑞. 基于CNN的点云图像融合目标检测[J]. 光电工程, 2021, 48(5): 200418. DOI: 10.12086/oee.2021.200418.
[21]郝佳, 姚国英, 周剑, 等. 基于图像和点云融合的三维小目标检测方法[J]. 测绘通报, 2025(3): 33-38. DOI: 10.13474/j.cnki.11-2246.2025.0306.
[22]Cheng X J, Wang P, Yang R G. Depth estimation via affinity learned with convolutional spatial propagation network[C]//Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018:108-125. DOI: 10.1007/978-3-030-01270-0_7.
[23]邵延华, 张铎, 楚红雨, 等. 基于深度学习的YOLO目标检测综述[J]. 电子与信息学报, 2022, 44(10): 3697-3708. DOI: 10.11999/JEIT210790.
[24]陈旭, 彭冬亮, 谷雨. 基于改进YOLOv5s的无人机图像实时目标检测[J]. 光电工程, 2022, 49(3): 210372. DOI: 10.12086/oee.2022.210372.
[25]程换新, 徐皓天, 骆晓玲. 基于改进YOLOv7的自动驾驶目标检测方法[J]. 激光杂志, 2024, 45(7): 91-96. DOI: 10.14016/j.cnki.jgzz.2024.07.091.
[26]张琴, 郭为安. 深度学习小目标检测算法综述[J]. 计算机应用研究, 2025, 42(10): 2893-2904. DOI: 10.19734/j.issn.1001-3695.2025.03.0067.
[27]Linder T, Pfeiffer K Y, Vaskevicius N, et al. Accurate detection and 3D localization of humans using a novel YOLO-based RGB-D fusion approach and synthetic training data[C]//2020 IEEE International Conference on Robotics and Automation (ICRA). Piscataway, NJ: IEEE, 2020: 1000-1006. DOI: 10.1109/ICRA40945.2020.9196899.
[28]Wang Y, Zell A. Yolo+FPN: 2D and 3D fused object detection with an RGB-D camera[C]//2020 25th International Conference on Pattern Recognition (ICPR). Los Alamitos, CA: IEEE Computer Society, 2021: 4657-4664. DOI: 10.1109/ICPR48806.2021.9413066.
[29]Sapkota R, Cheppally R H, Sharda A, et al. YOLO26: key architectural enhancements and performance benchmarking for real-time object detection[PP/OL].V5.arXiv(2026-03-16)[2026-06-28].https://doi.org/10.48550/arXiv.2509.25164.
[30]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. DOI: 10.1177/0278364913491297.
[31]Tian Y J, Ye Q X, Doermann D. YOLOv12: attention-centric real-time object detectors[C]//Advances in Neural Information Processing Systems38 (NeurIPS 2025).Red Hork, NY: Curran Associates Inc.,2025: 78433-78457. DOI: 10.52202/085713-2627.
[32]Hidayatullah P, Syakrani N, Sholahuddin M R, et al. YOLOv8 to YOLO11: a comprehensive architecture in-depth comparative review[PP/OL]. V4. arXiv(2026-05-04)[2026-06-28].https://doi.org/10.48550/arXiv.2501.13400.
[33]Wang A, Chen H, Liu L H, et al. YOLOv10: real-time end-to-end object detection[C]//Advances in Neural Information Processing Systems37(NeurIPS 2024). Red Hork, NY: Curran Associates Inc., 2024: 107984-108011. DOI: 10.52202/079017-3429.
[34]Wang C Y, Yeh I H, Liao H Y M. YOLOv9: learning what you want to learn using programmable gradient information[C]//Computer Vision-ECCV 2024. Cham: Springer Nature Switzerland, 2024: 1-21. DOI: 10.1007/978-3-031-72751-1_1.
[35]Terven J, Córdova-Esparza D M, Romero-González J A. A comprehensive review of YOLO architectures in computer vision: from YOLOv1 to YOLOv8 and YOLO-NAS[J]. Machine Learning and Knowledge Extraction, 2023, 5(4): 1680-1716. DOI: 10.3390/make5040083.
[36]Wang C C, He W, Nie Y, et al. Gold-YOLO: efficient object detector via gather-and-distribute mechanism[C]//Advances in Neural Information Processing Systems 36 (NeurIPS 2023). Red Hark, NY: Curran Associates Inc., 2023: 51094-51112. DOI: 10.52202/075280-2224.
[37]Liu S H, Zha J L, Sun J, et al. EdgeYOLO: an edge-real-time object detector[C]//2023 42nd Chinese Control Conference (CCC). Piscataway, NJ: IEEE, 2023: 7507-7512. DOI: 10.23919/CCC58697.2023.10239786.
[38]Li H L, Li J, Wei H B, et al. Slim-neck by GSConv: a lightweight-design for real-time detector architectures[J]. Journal of Real-Time Image Processing, 2024, 21(3): 62. DOI: 10.1007/s11554-024-01436-6.
[39]Cheng T H, Song L, Ge YX, et al. YOLO-World: real-time open-vocabulary object detection[C]//2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2024: 16901-16911. DOI: 10.1109/CVPR52733.2024.01599.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed