Multi-View Fusion Object Detection Method Based on Lidar Point Cloud Projection

Fig. 7. Visualization of object detection results (missing detection situation). (a) 2D camera images of original scenes; (b) single-view object detection results; (c) multi-view fusion object detection results; (d) point cloud object detection results

Download full size

Fig. 8. Visualization of object detection results (false detection situation). (a) 2D camera images of original scenes; (b) single-view object detection results; (c) multi-view fusion object detection results; (d) point cloud object detection results

Download full size

Table 1. 3D detection performance comparison between proposed and existing methods for Car

View table

Table 1. 3D detection performance comparison between proposed and existing methods for Car

Method	AP under different levels /%			Average /%
Method	Easy	Moderate	Hard	Average /%
AVOD^［13］	73.59	65.78	58.38	65.92
VoxelNet^［10］	77.47	65.11	57.73	66.77
F-PointNet^［12］	81.20	70.39	62.19	71.26
PointRCNN^［9］	85.29	75.08	68.38	76.25
STD^［11］	86.61	77.63	76.06	80.10
BEVDetNet^［16］	85.94	77.86	72.00	78.60
Proposed	87.13	79.67	77.31	81.37

Table 2. 3D detection performance comparison between proposed and existing methods for Ped

View table

Table 2. 3D detection performance comparison between proposed and existing methods for Ped

Method	AP under different levels /%			Average /%
Method	Easy	Moderate	Hard	Average /%
AVOD^［13］	38.28	31.51	26.98	32.26
VoxelNet^［10］	39.48	33.69	31.51	34.89
F‑PointNet^［12］	51.21	44.89	40.23	45.44
PointRCNN^［9］	49.43	41.78	38.63	43.28
STD^［11］	53.08	44.24	41.97	46.43
BEVDetNet^［16］	54.49	44.50	42.36	47.12
Proposed	53.27	48.16	46.59	49.34

Table 3. 3D detection performance comparison between proposed and existing methods for Cyc

View table

Table 3. 3D detection performance comparison between proposed and existing methods for Cyc

Method	AP under different levels /%			Average /%
Method	Easy	Moderate	Hard	Average /%
AVOD^［13］	60.11	44.90	38.80	47.94
VoxelNet^［10］	61.22	48.36	44.37	51.32
F‑PointNet^［12］	71.95	56.77	50.39	59.70
PointRCNN^［9］	73.93	59.60	53.59	62.37
STD^［11］	78.89	62.53	55.77	65.73
BEVDetNet^［16］	78.58	62.73	57.74	66.35
Proposed	78.49	64.18	61.24	67.97

Table 4. Comparison of time efficiency
View table
Table 4. Comparison of time efficiency
Method Model training duration /h Speed /（frame/s）
PointRCNN^［9］ 191 10
Proposed 62 34

Table 5. Comparison of detection accuracy in ablation experiments
View table
Table 5. Comparison of detection accuracy in ablation experiments
Method BEV RV ECA Car /% Ped /% Cyc /% mAP /%
Complex-YOLOv4^［21］ √ - - 76.67 45.90 63.91 62.16
YOLOv5 - √ - 75.21 46.61 64.34 62.05
Proposed √ √ - 80.58 48.76 66.43 65.26
√ √ √ 81.37 49.34 67.97 66.23

Tools

Get Citation

Copy Citation Text

Mu Zhou, Haocheng Ran, Yong Wang, Nan Du. Multi-View Fusion Object Detection Method Based on Lidar Point Cloud Projection[J]. Acta Optica Sinica, 2025, 45(12): 1228006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: Nov. 19, 2024

Accepted: Jan. 2, 2025

Published Online: May. 16, 2025

The Author Email: Mu Zhou (zhoumu@cqupt.edu.cn)

DOI:10.3788/AOS241772

CSTR:32393.14.AOS241772

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. 3D detection performance comparison between proposed and existing methods for Car

Table 1. 3D detection performance comparison between proposed and existing methods for Car

Table 2. 3D detection performance comparison between proposed and existing methods for Ped

Table 2. 3D detection performance comparison between proposed and existing methods for Ped

Table 3. 3D detection performance comparison between proposed and existing methods for Cyc

Table 3. 3D detection performance comparison between proposed and existing methods for Cyc

Table 4. Comparison of time efficiency

Table 4. Comparison of time efficiency

Table 5. Comparison of detection accuracy in ablation experiments

Table 5. Comparison of detection accuracy in ablation experiments