基于特征增稳的混合固态激光雷达目标检测

doi:10.19562/j.chinasae.qcgc.2024.06.008

摘要/Abstract

摘要：

稳定的点云特征提取对激光雷达三维目标检测至关重要。针对现有深度学习算法仅能处理机械旋转式激光雷达点云数据，而对混合固态激光雷达数据支持不足的问题，本文基于IA-SSD搭建了适用于混合固态激光雷达的目标检测模型。首先，在点云编码前端添加了Cloth Simulation Filtering （CSF）用于地面次优点过滤；其次，利用局部特征融合和全局双线性正则化层组成的注意力机制，从局部和全局共同推动点云几何信息与特征信息融合；再次，利用GhostGConv替换原有低效的逐点卷积，通过通道混洗机制加强点云的特征交互，构建了增强型特征提取网络。最后，在点检测器IA-SSD中整合了上述模块完成模型构建。在混合固态激光雷达数据集SimoSet开展的验证结果表明，所提出的方法显著优于SimoSet数据集支持的其他算法精度指标；在KITTI数据集中等难度检测中，所提方法将IA-SSD三分类平均检测精度分别提升了1.17、1.47、0.5百分点。

关键词: 车辆工程, 环境感知, 目标检测, 混合固态激光雷达, 特征增稳

Abstract:

Stable point cloud feature extraction is crucial for 3D object detection using LiDAR. For the limitation of existing deep learning algorithms that can only handle point clouds from mechanically rotating LiDAR but lack support for semi solid-state LiDAR data， in this paper a target detection model suitable for semi solid-state LiDAR based on IA-SSD is established. Firstly， Cloth Simulation Filtering （CSF） is added at the front end of point cloud encoding for ground suboptimal filtering. Secondly， an attention mechanism composed of local feature fusion and global bilinear regularization layers is utilized to promote the fusion of geometric information and feature information of point clouds from both local and global perspectives. Thirdly， GhostGConv is employed to replace the original inefficient point-by-point convolution， and enhanced feature interaction of point clouds is achieved through channel shuffling mechanism to construct an enhanced feature extraction network. Finally， the above modules are integrated into the point detector IA-SSD to complete model construction. The validation results conducted on the SimoSet semi solid-state lidar dataset show that the proposed method significantly outperforms other algorithms supported by the SimoSet dataset in terms of accuracy metrics. In the medium difficulty detection tasks on the KITTI dataset， the proposed method enhances the average detection accuracy of IA-SSD in the three categories by 1.17， 1.47， and 0.5 percentage points， respectively.

Key words: vehicle engineering, environmental perception, object detection, semi solid-state LiDAR, feature stability enhancement

金立生,张洪瑜,郭柏苍. 基于特征增稳的混合固态激光雷达目标检测[J]. 汽车工程, 2024, 46(6): 1015-1024.

Lisheng Jin,Hongyu Zhang,Baicang Guo. Semi Solid-State LiDAR Object Detection Algorithm Enhanced by Feature Stability Enhancement[J]. Automotive Engineering, 2024, 46(6): 1015-1024.

图/表 10

图1

图2

图3

图4

图5

图6

表1

图7

表2

表3

参考文献 27

1	HOSS M， SCHOLTES M， ECKSTEIN L. A review of testing object-based environment perception for safe automated driving［J］. Automotive Innovation， 2022， 5（3）： 223-250.
2	徐国艳，牛欢，郭宸阳，等. 基于三维激光点云的目标识别与跟踪研究［J］. 汽车工程， 2020， 42（1）： 38-46.
	XU Guoyan， NIU Huan， GUO Chenyang， et al. Research on target recognition and tracking based on 3D laser point cloud［J］. Automotive Engineering， 2020， 42（1）： 38-46.
3	SUN X， JIN L， HE Y， et al. SimoSet： a 3D object detection dataset collected from vehicle hybrid solid-state LiDAR［J］. Electronics， 2023， 12（11）： 2424.
4	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， 2018： 4490-4499.
5	YAN Y， MAO Y， LI B. Second： sparsely embedded convolutional detection［J］. Sensors， 2018， 18（10）： 3337.
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， 2019： 12697-12705.
7	SHI S， GUO C， JIANG L， et al. PV-RCNN： point-voxel feature set abstraction for 3D object detection［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2020： 10529-10538.
8	YIN T， ZHOU X， KRAHENBUHL P. Center-based 3D object detection and tracking［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2021： 11784-11793.
9	HE C， ZENG H， HUANG J， et al. Structure aware single-stage 3D object detection from point cloud［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2020： 11873-11882.
10	FAN L， WANG F， WANG N， et al. Fully sparse 3D object detection［J］. Advances in Neural Information Processing Systems， 2022， 35： 351-363.
11	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， 2023： 21674-21683.
12	LIANG A， ZHANG H， HUA H， et al. SPSNet： boosting 3D point-based object detectors with stable point sampling［J］. Engineering Applications of Artificial Intelligence， 2023， 126： 106807.
13	QI C R， SU H， MO K， et al. PointNet： deep learning on point sets for 3D classification and segmentation［C］. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition， 2017： 652-660.
14	QI C R， YI L， SU H， et al. PointNet++： deep hierarchical feature learning on point sets in a metric space［J］. Advances in Neural Information Processing Systems， 2017， 30.
15	QI C R， LITANY O， HE K， et al. Deep hough voting for 3D object detection in point clouds［C］. Proceedings of the IEEE/CVF International Conference on Computer Vision， 2019： 9277-9286.
16	YANG Z， SUN Y， 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， 2020： 11040-11048.
17	ZHANG Y， HU Q， XU G， et al. Not all points are equal： learning highly efficient point-based detectors for 3D lidar point clouds［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2022： 18953-18962.
18	HE Y， JIN L， GUO B， et al. Density-based road segmentation algorithm for point cloud collected by roadside LiDAR［J］. Automotive Innovation， 2023， 6（1）： 116-130.
19	ZHANG W， QI J， WAN P， et al. An easy-to-use airborne LiDAR data filtering method based on cloth simulation［J］. Remote Sensing， 2016， 8（6）： 501.
20	QIU S， ANWAR S， BARNES N. PnP-3D： a plug-and-play for 3D point clouds［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2021， 45（1）： 1312-1319.
21	HAN K， WANG Y， TIAN Q， et al. GhostNet： more features from cheap operations［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2020： 1580-1589.
22	ZHANG X， ZHOU X， LIN M， et al. ShuffleNet： an extremely efficient convolutional neural network for mobile devices［C］. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition， 2018： 6848-6856.
23	SCHINAGL D， KRISPEL G， POSSEGGER H， et al. OccAM's laser： occlusion-based attribution maps for 3D object detectors on LiDAR data［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2022： 1141-1150.
24	SHI S， WANG X， LI H. PointrCNN： 3D object proposal generation and detection from point cloud［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2019： 770-779.
25	SHI W， 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， 2020： 1711-1719.
26	LIU Z， ZHAO X， HUANG T， et al. TANet： robust 3D object detection from point clouds with triple attention［J］. Proceedings of the AAAI Conference on Artificial Intelligence， 2020， 34（7）： 11677-11684.
27	SHI S， WANG Z， WANG X， et al. Part-A ² Net： 3D part-aware and aggregation neural network for object detection from point cloud［J］. arXiv preprint arXiv：， 2019.

算法	范式	类别	AP_3D/%
算法	范式	类别	LEVEL_1	LEVEL_2
SECOND	体素	Car	80.23	65.71
		Cyc	81.52	45.08
		Ped	78.16	38.92
PointPillars	体素	Car	78.73	63.15
		Cyc	84.05	47.13
		Ped	81.50	42.55
SA-SSD	体素	Car	82.79	68.93
		Cyc	86.36	47.87
		Ped	85.28	45.29
Point-RCNN	点	Car	80.65	67.00
		Cyc	85.27	47.53
		Ped	85.45	21.87
Our	点	Car	84.53	70.65
		Cyc	86.68	48.23
		Ped	85.85	45.24

算法	年份	3D Car （IoU=0.7）			3D Ped（IoU=0.5）			3D Cyc （IoU=0.5）
算法	年份	Easy	Mod	Hard	Easy	Mod	Hard	Easy	Mod	Hard
VoxelNet^［4］	2018	77.47	65.11	57.73	39.48	33.69	31.5	61.22	48.36	44.37
SECOND^［5］	2018	84.65	75.96	68.71	45.31	35.52	33.14	75.83	60.82	53.67
PointPillars^［6］	2019	82.58	74.31	68.99	51.45	41.92	38.89	77.10	58.65	51.92
Point-GNN^［25］	2020	88.33	79.47	72.29	51.92	43.77	40.10	78.60	63.48	57.08
PointRCNN^［24］	2020	86.96	75.64	70.70	51.92	43.77	40.10	78.60	63.48	57.08
TANet^［26］	2020	84.39	75.94	68.82	53.72	44.34	40.49	75.70	59.44	52.53
Part-A^2［27］	2020	87.81	78.49	73.51	53.10	43.35	40.06	79.17	63.52	56.93
IA-SSD^［17］	2022	88.87	80.32	75.10	49.01	41.20	38.03	80.78	66.01	58.12
Ours		89.82	81.49	75.78	50.88	42.67	39.82	81.21	66.51	58.43

算法	AP_3D/% （LEVEL_1）
算法	Car	Cyc	Ped
IA-SSD	82.41	85.46	85.28
+CSF	83.32	85.94	85.51
+LCF&GBR	84.50	86.67	85.84
+GhostGConv	84.41	86.56	85.75
+通道混洗	84.53	86.68	85.85

[1]	朱冰,唐衍鹏,张东波,赵健,陈志成. 考虑多维非线性扰动的电子助力制动系统自适应压力控制[J]. 汽车工程, 2024, 46(6): 1096-1103.
[2]	邓云红,赵治国,杨一飞,于勤. ADAS系统视觉与毫米波雷达分布式抗差卡尔曼滤波融合算法[J]. 汽车工程, 2024, 46(5): 805-815.
[3]	程腾,倪昊,张强,王文冲,石琴. 基于虚拟点云的二阶段多模态融合网络[J]. 汽车工程, 2024, 46(2): 222-229.
[4]	马雷, 杨顺清, 王欢欢, 翟家琛, 徐健傲. 融合图像显著性特征的轻量级目标检测算法[J]. 汽车工程, 2024, 46(1): 84-91.
[5]	马天飞,李波,朱冰,赵健. 考虑大气条件影响的自动泊车系统超声波雷达建模[J]. 汽车工程, 2023, 45(9): 1646-1654.
[6]	赵东宇, 赵树恩. 基于级联YOLOv7的自动驾驶三维目标检测[J]. 汽车工程, 2023, 45(7): 1112-1122.
[7]	韩勇,林旭洁,黄红武,蔡鸿瑜,罗金镕,李燕婷. 典型汽车碰撞事故场景中行人运动轨迹预测方法[J]. 汽车工程, 2023, 45(6): 1022-1030.
[8]	赵嘉豪,齐志权,齐智峰,王皓,何磊. 基于轮胎特征点的并行大型车辆朝向角计算[J]. 汽车工程, 2023, 45(6): 1031-1039.
[9]	陈妍妍,王海,蔡英凤,陈龙,李祎承. 基于检测的高效自动驾驶实例分割方法[J]. 汽车工程, 2023, 45(4): 541-550.
[10]	刘正发,吴亚,刘佩根,顾荣琦,陈广. 基于特征和标签联合分布匹配的智能驾驶跨域自适应目标检测[J]. 汽车工程, 2023, 45(11): 2082-2091.
[11]	胡杰,徐博远,熊宗权,昌敏杰,郭迪,谢礼浩. 基于多尺度掩码分类域自适应网络的跨域目标检测算法[J]. 汽车工程, 2022, 44(9): 1327-1338.
[12]	金立生,贺阳,王欢欢,霍震,谢宪毅,郭柏苍. 基于自适应阈值DBSCAN的路侧点云分割算法[J]. 汽车工程, 2022, 44(7): 987-996.
[13]	李希宸,张虹,解晓天,汪博文,王新宇. 针对ASR系统防误触的复杂路面识别算法[J]. 汽车工程, 2022, 44(5): 756-763.
[14]	谢德胜,徐友春,陆峰,潘世举. 基于多传感器信息融合的3维目标实时检测[J]. 汽车工程, 2022, 44(3): 340-349.
[15]	刘子龙,沈祥飞. 融合Lite-HRNet的Yolo v5双模态自动驾驶小目标检测方法[J]. 汽车工程, 2022, 44(10): 1511-1520.