融合密集连接和高斯距离的三维目标检测算法

doi:10.13229/j.cnki.jdxbgxb.20230105

吉林大学学报(工学版) ›› 2024, Vol. 54 ›› Issue (12): 3589-3600.doi: 10.13229/j.cnki.jdxbgxb.20230105

融合密集连接和高斯距离的三维目标检测算法

程鑫^1,²(),刘升贤¹,周经美³(),周洲¹,赵祥模¹

^1.长安大学信息工程学院，西安 710018
^2.公安部交通管理科学研究所，江苏无锡 214151
^3.长安大学电子与控制工程学院，西安 710018

收稿日期:2023-02-06 出版日期:2024-12-01 发布日期:2025-01-24
通讯作者: 周经美 E-mail:xincheng@chd.edu.cn;jmzhou@chd.edu.cn
作者简介:程鑫（1990-），男，副教授，博士.研究方向：车路协同与深度学习.E-mail：xincheng@chd.edu.cn
基金资助:
国家自然科学基金项目(52472337);中国博士后科学基金项目(2023T160129);交通运输部重点科技项目(2022-ZD6-079);陕西省重点研发计划项目(2022QCY-LL-29);云南省交科院科技研发项目(JKYZLX-2023-12)

3D object detection algorithm fusing dense connectivity and Gaussian distance

Xin CHENG^1,²(),Sheng-xian LIU¹,Jing-mei ZHOU³(),Zhou ZHOU¹,Xiang-mo ZHAO¹

^1.School of Information Engineering，Chang'an University，Xi'an 710018，China
^2.Traffic Management Research Institute of the Ministry of Public Security，Wuxi 214151，China
^3.School of Electronics and Control Engineering，Chang'an University，Xi'an 710018，China

Received:2023-02-06 Online:2024-12-01 Published:2025-01-24
Contact: Jing-mei ZHOU E-mail:xincheng@chd.edu.cn;jmzhou@chd.edu.cn

摘要/Abstract

摘要：

为加强对小目标的感知，在F-PointNet网络的基础上，结合密集连接方法和高斯距离特征，提出了FDG-PointNet三维目标检测模型，融合高斯距离特征作为附加注意力特征，有效改善了F-PointNet网络实例分割准确率不高的问题，增强了对点云视锥体中的噪声的过滤；基于密集连接可以加强特征提取的特点，改进主干特征提取PointNet++网络，加强点云特征重用，缓解特征提取过程中对小目标的特征提取程度过低与梯度消失问题，提高三维目标边界框回归的准确性。研究结果表明：本文算法在简单、中等、困难三个难度等级下对汽车、行人、骑行人3种类别的检测整体优于基准方法F-PointNet，在中等难度下对汽车、行人、骑行人的检测分别取得71.12%、61.23%、55.71%的平均检测精度，其中对行人检测提升最明显，在简单和中等难度下提升幅度分别达5.5%和3.1%。综上所述，本文的FDG-PointNet算法有效解决了F-PointNet中小物体检测的低准确性问题，具有较强的适用性。

关键词: 计算机科学与技术, 三维目标检测, 激光雷达, 密集连接, 高斯距离

Abstract:

To enhance the perception of small objects， based on the F-PointNet network，the FDG-PointNet 3D object detection model is proposed by combining dense connection and Gaussian distance features. Gaussian distance features is fused as additional attention features， and it effectively solves the low accuracy of instance segmentation in the F-PointNet network and enhances the noise filtering in the point cloud view cone. Based on the characteristics that dense connection can enhance feature extraction， the dense connection is used to improve PointNet++ network and enhance point cloud feature reuse. It alleviates low degree of feature extraction and gradient disappearance for small objects in the feature extraction process， and improves the accuracy of 3D object bounding box regression. The experimental results show that the proposed algorithm outperforms the benchmark method F-PointNet in three levels （easy， moderate， and hard） for the detection of car， pedestrian， and cyclist， which can achieve the average detection accuracy of 71.12%， 61.23%， and 55.71% for car， pedestrian， and cyclist at moderate level. It has the most significant improvement for pedestrian detection， and can increase 5.5% and 3.1% at easy and moderate levels， respectively. In summary，compared to F-PointNet algorithm， the proposed FDG-PointNet algorithm effectively solves the low accuracy of small objects detection and has strong applicability.

Key words: computer science and technology, 3D object detection, lidar, dense connectivity, gaussian distance

中图分类号:

TP391.4

程鑫,刘升贤,周经美,周洲,赵祥模. 融合密集连接和高斯距离的三维目标检测算法[J]. 吉林大学学报(工学版), 2024, 54(12): 3589-3600.

Xin CHENG,Sheng-xian LIU,Jing-mei ZHOU,Zhou ZHOU,Xiang-mo ZHAO. 3D object detection algorithm fusing dense connectivity and Gaussian distance[J]. Journal of Jilin University(Engineering and Technology Edition), 2024, 54(12): 3589-3600.

图/表 13

图1

图2

图3

图4

图5

表1

表2

表3

表4

表5

表6

表7

图6

参考文献 28

1	Cheng X, Zhou J M, Liu P Y, et al. 3D vehicle object tracking algorithm based on bounding box similarity measurement[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(12): 15844-15854.
2	Cheng X, Zhou J M, Song J C, et al. A highway traffic image enhancement algorithm based on improved GAN in complex weather conditions[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(8): 8716-8726.
3	Yurtsever E, Lambert J, Carballo A, et al. A survey of autonomous driving: common practices and emerging technologies[J]. IEEE Access, 2020, 8: 58443-58469.
4	Cui Y D, Chen R, Chu W B, et al. Deep learning for image and point cloud fusion in autonomous driving: a review[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(2): 722-739.
5	Zhou Y, Tuzel O. VoxelNet: end-to-end learning for point cloud based 3D object detection[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 4490-4499.
6	Lang A H, Vora S, Caesar H, et al. PointPillars: fast encoders for object detection from point clouds [C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 12697-12705.
7	Yan Y, Mao Y X, Li B. SECOND: sparsely embedded convolutional detection[J]. Sensors, 2018, 18(10): 3337.
8	Deng J J, Shi S S, Li P W, et al. Voxel R-CNN: towards high performance voxel-based 3D object detection[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35(2): 1201-1209.
9	Li B. 3D fully convolutional network for vehicle detection in point cloud[C]∥2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Vancouver, Canada, 2017: 1513-1518.
10	程鑫, 王宏飞, 周经美, 等. 基于体素柱形的激光雷达点云车辆目标检测算法[J]. 中国公路学报, 2023, 36(3): 247-260.
	Cheng Xin, Wang Hong-fei, Zhou Jing-mei, et al. Vehicle detection algorithm based on voxel pillars from lidar point clouds[J]. China Journal of Highway and Transport, 2023, 36(3): 247-260.
11	Qi C R, Su H, Mo K, et al. PointNet: deep learning on point sets for 3D classification and segmentation[C]∥2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, USA, 2017: 77-85.
12	Qi C R, Yi L, Su H, et al. PointNet++: deep hierarchical feature learning on point sets in a metric space[C]∥Advances in Neural Information Processing Systems, Long Beach, USA, 2017: 5105 - 5114 .
13	Shi S S, Wang X G, Li H S. PointRCNN: 3D object proposal generation and detection from point cloud[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 770-779.
14	Yang Z T, Sun Y A, Liu S, et al. 3DSSD: point-based 3D single stage object detector[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 11040-11048.
15	Yang Z T, Sun Y A, Liu S, et al. STD: sparse-to-dense 3D object detector for point cloud[C]∥Proceedings of the IEEE/CVF International Conference on Computer Vision, Long Beach, USA, 2019: 1951-1960.
16	Shi W J, Rajkumar R. Point-GNN: graph neural network for 3D object detection in a point cloud[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle,USA, 2020: 1711-1719.
17	Yin T W, Zhou X Y, Krahenbuhl P. Center-based 3D object detection and tracking[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 11784-11793.
18	Chen X Z, Ma H M, Wan J, et al. Multi-view 3D object detection network for autonomous driving[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 1907-1915.
19	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), Madrid, Spain, 2018: No.8594049.
20	Qi C R, Liu W, Wu C X, et al. Frustum PointNets for 3D object detection from RGB-D data[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 918-927.
21	Vora S, Lang A H, Helou B, et al. PointPainting: sequential fusion for 3D object detection[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA, 2020: 4604-4612.
22	Xu D F, Anguelov D, Jain A. PointFusion: deep sensor fusion for 3D bounding box estimation[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 244-253.
23	Wang Z X, Jia K. Frustum ConvNet: sliding frustums to aggregate local point-wise features for amodal 3D object detection[C]∥2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Macau, China, 2019: 1742-1749.
24	Huang G, Liu Z, Maaten L, et al. Densely connected convolutional networks[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 4700-4708.
25	Geiger A, Lenz P, Stiller C, et al. Vision meets robotics: the KITTI dataset[J]. The International Journal of Robotics Research, 2013, 32(11): 1231-1237.
26	Klokov R, Lempitsky V. Escape from cells: deep kd-networks for the recognition of 3D point cloud models[C]∥Proceedings of the IEEE International Conference on Computer Vision, Honolulu, USA, 2017: 863-872.
27	Li J X, Chen B M, Lee G H. SO-Net: self-organizing network for point cloud analysis[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 9397-9406.
28	Graham B, Engelcke M, Maaten L. 3D semantic segmentation with submanifold sparse convolutional networks[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 9224-9232.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

方法	输入	总体精度OA/%
Subvolume	voxels	89.2
PointNet^［11］	point cloud	89.2
Kd-net^［26］	point cloud	90.6
SO-net^［27］	point cloud	90.9
PointNet++^［12］	point cloud	90.7
本文	point cloud	91.3

方法	mean	aero	bag	cap	car	chair	ear phone	guitar	knife	lamp	laptop	motor	mug	pistol	rocket	skate board	table
PointNe^t［11］	83.7	83.4	78.7	82.5	74.9	89.6	73.0	91.5	85.9	80.8	95.3	65.2	93.0	81.2	57.9	72.8	80.6
SSCN^［28］	84.7	81.6	81.7	81.9	75.2	90.2	74.9	93.0	86.1	84.7	95.6	66.7	92.7	81.6	60.6	82.9	82.1
PointNet++^［12］	85.1	82.4	79.0	87.7	77.3	90.8	71.8	91.0	85.9	83.7	95.3	71.6	94.1	81.3	58.7	76.4	82.6
本文	85.3	82.6	82.6	86.3	78.1	91.0	71.1	90.8	87.6	83.1	96.0	72.0	94.4	81.0	58.2	76.8	81.6

模型	汽车			行人			骑行人
模型	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
F-PointNet（v1）	87.82	82.44	74.50	70.38	62.07	56.77	76.72	57.71	52.88
本文	87.90	82.24	74.53	71.61	65.07	57.39	78.55	58.45	54.95

模型	汽车			行人			骑行人
模型	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
F-PointNet（v1）	83.26	69.28	62.56	66.04	57.05	49.84	70.52	51.50	50.95
本文	83.83	69.73	63.10	67.08	57.90	50.90	71.13	52.59	49.40

密集连接	高斯距离	汽车			行人			骑行人
密集连接	高斯距离	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
—	—	83.76	70.92	63.65	64.75	58.11	51.23	73.15	54.89	51.62
√	—	81.18	71.34	63.82	68.15	59.84	52.47	75.03	55.99	52.74
—	√	83.81	70.64	63.77	66.99	59.10	52.25	76.85	55.81	51.67
√	√	84.39	71.12	64.28	70.21	61.23	53.27	75.37	55.71	51.71

融合密集连接和高斯距离的三维目标检测算法

3D object detection algorithm fusing dense connectivity and Gaussian distance

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 28

相关文章 6

Metrics

本文评价

推荐阅读 10

[1]	陶博,颜伏伍,尹智帅,武冬梅. 基于高精度地图增强的三维目标检测算法[J]. 吉林大学学报(工学版), 2023, 53(3): 802-809.
[2]	李雪梅,王春阳,刘雪莲,谢达. 基于SESTH的线性调频连续波激光雷达信号时延估计[J]. 吉林大学学报(工学版), 2022, 52(4): 950-958.
[3]	院老虎,连冬杉,张亮,刘义. 基于密集连接卷积网络和支持向量机的飞行器机械部件故障诊断[J]. 吉林大学学报(工学版), 2021, 51(5): 1635-1641.
[4]	宗长富,文龙,何磊. 基于欧几里得聚类算法的三维激光雷达障碍物检测技术[J]. 吉林大学学报(工学版), 2020, 50(1): 107-113.
[5]	潘海阳, 刘顺安, 姚永明. 基于深度信息的自主空中加油技术[J]. 吉林大学学报(工学版), 2014, 44(6): 1750-1756.
[6]	尚涛, 贺会超, 王昕, 宫文斌, 戴永强. 二维激光雷达实现三维成像的算法[J]. 吉林大学学报(工学版), 2012, 42(增刊1): 91-95.

模型	汽车			行人			骑行人
模型	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
MV3D（BV+FV）^［18］	71.19	56.60	55.30	N/A	N/A	N/A	N/A	N/A	N/A
MV3D（BV+FV+RGB）^［18］	71.29	62.68	56.56	N/A	N/A	N/A	N/A	N/A	N/A
RT3D^［29］	72.85	61.64	64.38	N/A	N/A	N/A	N/A	N/A	N/A
PointFusion^［22］	77.92	63.00	53.27	33.36	28.04	23.38	49.34	29.42	26.98
VoxelNet^［5］	81.97	65.46	62.85	57.86	53.42	48.87	67.17	47.65	45.11
F-PointNet（v1）^［20］	83.26	69.28	62.56	66.04	57.05	49.84	70.52	51.50	50.95
F-PointNet（v2）^［20］	83.76	70.92	63.65	64.75	58.11	51.23	73.15	54.89	51.62
本文	84.39	71.12	64.28	70.21	61.23	53.27	75.37	55.71	51.71