Three-Dimensional Object Detection Technology Based on Point Cloud Data

Fig. 3. Comparision of point cloud feature extraction methods. (a) Voxel-based point cloud feature extraction method; (b) point-based point cloud feature extraction method; (c) graph-based point cloud feature extraction method

Download full size

Fig. 4. Milestone timeline of 3D object detection in point clouds

Download full size

Fig. 5. Pipeline of voxel-based object detection method

Download full size

Fig. 6. Pipeline of point-based object detection method

Download full size

Fig. 7. Pipeline of graph-based object detection method

Download full size

Fig. 8. Samples of outdoor 3D object detection datasets. (a) KITTI; (b) Waymo; (c) nuScenes; (d) STCrowd

Download full size

Table 1. Comparison of commonly used 3D object detection datasets

View table

Table 1. Comparison of commonly used 3D object detection datasets

Dataset	Scene	Year	Data type	3D bounding box	Category
KITTI^［7］	Outdoor	2012	Point cloud +image	2×10⁵	8
Waymo^［61］	Outdoor	2019	Point cloud +image	1.2×10⁶	4
nuScenes^［62］	Outdoor	2019	Point cloud +image	4×10⁵	23
STCrowd^［63］	Outdoor	2022	Point cloud +image	2.19×10⁵	1
NYU-Depth^［64］	Indoor	2012	Image+depth map	3.5×10⁴	40
SUN3D^［65］	Indoor	2013	Image+depth map	—	—
SUN RGB-D^［66］	Indoor	2015	Image+depth map	5.8×10⁴	800
ScanNet^［67］	Indoor	2017	Image+depth map	—	—

Table 2. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on KITTI dataset

View table

Table 2. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on KITTI dataset

Type	Method	Car			Pedestrian			Cyclist
Type	Method	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard
Voxel-based	VoxelNet^［16］	77.47	65.11	57.73	39.48	33.69	31.50	61.22	48.36	44.37
	SECOND^［17］	84.65	75.96	68.71	45.31	35.52	33.14	75.83	60.82	53.67
	PointPillars^［18］	82.58	74.31	68.99	51.45	41.92	38.89	77.10	58.65	51.92
	PartA^2^［70］	87.81	78.49	73.51	53.10	43.35	40.06	79.17	63.52	56.93
	TANet^［71］	84.39	75.94	68.82	53.72	44.34	40.49	75.70	59.44	52.53
	SegVoxelNet^［72］	86.04	76.13	70.76	—	—	—	—	—	—
	CIA-SSD^［21］	89.59	80.28	72.87	—	—	—	—	—	—
	Voxel R-CNN^［19］	90.90	81.62	77.06	—	—	—	—	—	—
	SE-SSD^［22］	91.49	82.54	77.15	—	—	—	—	—	—
	VoTr-TSD^［25］	89.90	82.09	79.14	—	—	—	—	—	—
	CT3D^［26］	87.83	81.77	77.16	—	—	—	—	—	—
	VoxSeT^［27］	88.53	82.06	77.46	—	—	—	—	—	—
Point-based	Point R-CNN^［38］	86.96	75.64	70.70	47.98	39.37	36.01	74.96	58.82	52.53
	3DSSD^［39］	88.36	79.57	74.55	54.64	44.27	40.23	82.48	64.10	56.90
	IA-SSD（single）^［40］	88.87	80.32	75.10	47.90	41.03	37.98	82.36	66.25	59.70
	IA-SSD（multi）^［40］	88.34	80.13	75.04	46.51	39.03	35.61	78.35	61.94	55.70
	SASA^［41］	88.76	82.16	77.16	—	—	—	—	—	—
Graph-based	Point-GNN^［42］	88.33	79.47	72.29	51.92	43.77	40.14	78.6	63.48	57.08
	PC R-GNN^［43］	89.13	79.90	75.54	—	—	—	—	—	—
	GraR-Vol^［44］	91.89	83.27	77.78	—	—	—	—	—	—
	GraR-Po^［44］	91.79	83.18	77.98	—	—	—	—	—	—
	GraR-Vo^［44］	91.29	82.77	77.20	—	—	—	—	—	—
	GraR-Pi^［44］	90.94	82.42	77.00	—	—	—	—	—	—
Voxel+point-based	FP R-CNN^［45］	85.29	77.40	70.24	—	—	—	—	—	—
	STD^［46］	87.95	79.71	75.09	53.29	42.47	38.35	78.69	61.59	55.30
	PV R-CNN^［47］	90.25	81.43	76.82	52.17	43.29	40.29	78.60	63.71	57.65
	SA-SSD^［48］	88.75	79.79	74.16	—	—	—	—	—	—
	ImpDet^［73］	88.39	82.14	76.98	—	—	—	—	—	—
Multimode-based	MV3D^［52］	74.97	63.63	54.00	—	—	—	—	—	—
	F-PointNet^［50］	82.19	69.79	60.59	50.53	42.15	38.08	72.27	56.12	49.01
	AVOD^［53］	76.39	66.47	60.23	36.10	27.86	25.76	57.19	42.08	38.29
	ContFuse^［58］	83.68	68.78	61.67	—	—	—	—	—	—
	MMF^［59］	88.40	77.43	70.22	—	—	—	—	—	—

Table 3. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on Waymo dataset

View table

Table 3. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on Waymo dataset

Level	Method	3D				BEV
Level	Method	Overall	0-30 m	30-50 m	50 m-Inf	Overall	0-30 m	30-50 m	50 m-Inf
LEVEL_1（IoU is 0.7）	PointPillars^［18］	56.62	81.01	51.75	27.94	75.57	92.10	74.06	55.47
	MVF^［74］	62.93	86.30	60.02	36.02	80.40	93.59	79.21	63.09
	PV R-CNN^［47］	70.30	91.92	69.21	42.17	82.96	97.35	82.99	64.97
	Pillar-OD^［75］	69.80	88.53	66.50	42.93	87.11	95.78	84.87	72.12
	Voxel R-CNN^［19］	75.59	92.49	74.09	53.15	88.19	97.62	87.34	77.70
	LiDAR R-CNN^［76］	76.00	92.10	74.60	54.50	90.10	97.00	89.50	78.90
	CenterPoint^［23］	76.86	92.27	75.31	54.10	91.61	97.19	91.05	82.06
	PVGNet^［77］	74.00	—	—	—	—	—	—	—
	VoTR-TSD^［25］	74.95	92.28	73.36	51.09	—	—	—	—
	CT3D^［26］	76.30	92.51	75.07	55.36	90.50	97.64	88.06	78.89
	Pyramid-PV^［78］	76.30	92.67	74.91	54.54	—	—	—	—
	VoxSeT^［27］	76.02	91.13	75.75	54.23	89.12	95.12	87.36	77.78
	GraR-Ce^［44］	80.77	93.59	79.68	60.41	92.69	97.56	92.15	84.13
	ImpDet^［73］	74.38	91.98	72.86	49.13	—	—	—	—
LEVEL_2（IoU is 0.7）	PV R-CNN^［47］	65.36	91.58	65.13	36.46	77.45	94.64	80.39	55.39
	Voxel R-CNN^［19］	66.59	91.74	67.89	40.80	81.07	96.99	81.37	63.26
	LiDAR R-CNN^［76］	68.30	91.30	68.50	42.40	81.70	94.30	82.30	65.80
	CenterPoint^［23］	69.09	91.41	69.43	42.40	85.43	96.35	86.44	70.06
	VoTR-TSD^［25］	65.91	—	—	—	—	—	—	—
	CT3D^［26］	69.04	91.76	68.93	42.60	81.74	97.05	82.22	64.34
	Pyramid-PV^［78］	67.23	—	—	—	—	—	—	—
	VoxSeT^［27］	68.16	91.03	67.13	42.23	76.13	94.13	81.78	58.13
	GraR-Ce^［44］	72.55	92.75	73.74	47.84	86.56	96.79	87.59	72.06

Tools

Get Citation

Copy Citation Text

Jianan Li, Ze Wang, Tingfa Xu. Three-Dimensional Object Detection Technology Based on Point Cloud Data[J]. Acta Optica Sinica, 2023, 43(15): 1515001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Mar. 29, 2023

Accepted: Jun. 5, 2023

Published Online: Aug. 3, 2023

The Author Email: Xu Tingfa (ciom_xtf1@bit.edu.cn)

DOI:10.3788/AOS230745

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparison of commonly used 3D object detection datasets

Table 1. Comparison of commonly used 3D object detection datasets

Table 2. Average precision (pAP) comparison of point cloud object detection methods on KITTI dataset

Table 2. Average precision (pAP) comparison of point cloud object detection methods on KITTI dataset

Table 3. Average precision (pAP) comparison of point cloud object detection methods on Waymo dataset

Table 3. Average precision (pAP) comparison of point cloud object detection methods on Waymo dataset

Table 2. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on KITTI dataset

Table 2. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on KITTI dataset

Table 3. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on Waymo dataset

Table 3. Average precision ( $p_{A P}$ ) comparison of point cloud object detection methods on Waymo dataset