3D Object Detection Based on Fusion of Voxel Texture Information and Deep Semantic Features

Longfei Wang; Likang Fan; Yiqiang Peng; Jie Cao; Liu He; Xulei Liu; Xiyuan Gao

doi:10.3788/LOP242537

Laser & Optoelectronics Progress, Volume. 62, Issue 16, 1615006(2025)

3D Object Detection Based on Fusion of Voxel Texture Information and Deep Semantic Features

Longfei Wang¹, Likang Fan^1,2,3、*, Yiqiang Peng^1,4,5, Jie Cao¹, Liu He^1,2,3, Xulei Liu^1,2,3, and Xiyuan Gao¹

Author Affiliations

¹School of Automobile and Transportation, Xihua University, Chengdu 610039, Sichuan , China

²Vehicle Measurement Control and Safety Key Laboratory of Sichuan Province, Xihua University, Chengdu 610039, Sichuan , China

³Provincial Engineering Research Center for New Energy Vehicle Intelligent Control and Simulation Test Technology of Sichuan, Chengdu 610039, Sichuan , China

⁴Yibin Institute in Xihua University, Yibin 644000, Sichuan , China

⁵Sichuan Intelligent and New Energy Automobile Industry College, Yibin 644000, Sichuan , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Flow chart of the Voxel-AESC algorithm. (a) 3D voxel processing unit; (b) BEV image feature processing unit

Download full size

Fig. 2. VFE module schematic diagram

Download full size

Fig. 3. ISC3D module schematic diagram

Download full size

Fig. 4. CASA module schematic diagram

Download full size

Fig. 5. Point cloud views and real scene images of the proposed algorithm on KITTI dataset

Download full size

Fig. 6. Real vehicle hardware platform. (a) Real vehicle hardware image; (b)(c) schematic diagrams of systems

Download full size

Fig. 7. Visualization results of the on-campus detection. (a) Camera perspective; (b) point cloud perspective; (c) visualization results

Download full size

Table 1. Comparison of 3D detection accuracy results of different algorithms in KITTI validation set

View table

Table 1. Comparison of 3D detection accuracy results of different algorithms in KITTI validation set

Method	Car			Pedestrian			Cyclist			Average
Method	Easy	Moderate	Hard	Easy	Moderate	Hard	Easy	Moderate	Hard	Average
Pointpillars	87.50	77.01	74.77	66.73	61.06	56.50	83.65	63.40	59.71	67.16
PointRCNN	89.01	78.77	78.10	62.69	55.36	51.60	84.48	65.37	59.83	66.50
Point-GNN	89.33	79.47	78.29	61.92	53.77	50.14	86.60	67.48	62.58	66.91
Part- $A^{2}$	89.56	79.41	78.84	65.69	60.05	55.45	85.50	68.90	64.53	69.45
PV-RCNN	91.54	82.67	80.24	60.39	53.14	48.49	88.05	70.99	66.54	68.93
3DSSD	88.36	79.40	74.55	64.64	44.27	40.23	82.48	64.10	56.90	62.59
VoxelNet	87.93	75.37	73.21	67.81	63.52	58.87	77.69	58.72	51.63	65.87
SECOND	88.16	78.18	77.04	56.00	50.02	43.64	79.96	63.43	56.67	63.88
Voxel-AESC	88.62	79.19	76.54	57.92	55.62	45.19	81.29	66.42	58.07	67.08
*	+0.46	+1.01	-0.50	+1.92	+5.60	+1.55	+1.33	+2.99	+1.40	+3.20

Table 2. Comparison of BEV detection accuracy results of different algorithms in KITTI validation set

View table

Table 2. Comparison of BEV detection accuracy results of different algorithms in KITTI validation set

Method	Car			Pedestrian			Cyclist				Average
Method	Easy	Moderate	Hard	Easy	Moderate	Hard		Easy	Moderate	Hard	Average
Pointpillars	90.07	86.56	82.81	57.60	48.64	45.78	79.90		62.73	55.58	65.98
PointRCNN	92.13	87.39	82.72	54.77	46.13	42.84	82.56		67.24	60.28	66.92
Point-GNN	93.11	89.17	83.90	55.36	47.07	44.61	85.04		67.62	61.14	67.95
Part- $A^{2}$	91.70	87.79	84.47	—	—	—	81.91		68.12	61.92	—
PV-RCNN	94.98	90.65	86.14	—	—	—	82.49		68.89	62.41	—
3DSSD	92.66	89.02	85.86	60.54	49.94	45.73	85.04		67.62	61.14	68.86
VoxelNet	89.35	79.26	77.39	46.13	40.74	38.11	66.70		54.76	50.55	58.25
SECOND	91.81	86.37	81.04	55.99	45.02	40.93	76.50		56.05	49.45	62.48
Voxel-AESC	92.50	87.99	86.99	57.61	52.29	47.79	84.77		67.83	63.11	69.37
*	+0.69	+1.62	+5.95	+1.62	+7.27	+6.86	+8.27		+11.78	+13.66	+6.22

Table 3. Validation of the effectiveness of ISC3D and CASA modules
View table
Table 3. Validation of the effectiveness of ISC3D and CASA modules
Module Average accuracy
Car Pedestrian Cyclist
None 78.13 51.13 63.33
ISC3D 78.78 52.88 64.98
CASA 78.36 54.45 65.13
ISC3D+CASA 79.19 55.62 66.42

Table 4. Hardware list of real vehicle platform
View table
Table 4. Hardware list of real vehicle platform
Equipment Number Brand Type
IPC 1 ADVANTECH 610 L
INS 1 CHCNAV 410
CDC 2 Freescale XEP100
GPS 2 CHCNAV 410
MMW radar 1 Continental ARS408
LiDAR 1 Leishen C32
Camera 2 Molose UC50

Table 5. Real vehicle data test results
View table
Table 5. Real vehicle data test results
Type Number Correct Wrong Omissive Accuracy /%
Pedestrian 25 13 6 6 52.00
Cyclist 22 13 5 4 59.09

Tools

Get Citation

Copy Citation Text

Longfei Wang, Likang Fan, Yiqiang Peng, Jie Cao, Liu He, Xulei Liu, Xiyuan Gao. 3D Object Detection Based on Fusion of Voxel Texture Information and Deep Semantic Features[J]. Laser & Optoelectronics Progress, 2025, 62(16): 1615006

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites