Three-Dimensional Reconstruction Methods for Obstacles in Complex Parking Scenarios

Shidian Ma; Yuxuan Huang; Haobin Jiang; Aoxue Li; Mu Han; Chenxu Li

doi:10.3788/LOP241025

Laser & Optoelectronics Progress, Volume. 61, Issue 24, 2415008(2024)

Three-Dimensional Reconstruction Methods for Obstacles in Complex Parking Scenarios

Shidian Ma^1、*, Yuxuan Huang¹, Haobin Jiang¹, Aoxue Li², Mu Han³, and Chenxu Li²

Author Affiliations

¹Automotive Engineering Research Institute, Jiangsu University, Zhenjiang 212013, Jiangsu , China

²School of Automotive and Traffic Engineering, Jiangsu University, Zhenjiang 212013, Jiangsu , China

³School of Computer Science and Communication Engineering, Jiangsu University, Zhenjiang 212013, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(23)

Fig. 1. U-Net model structure schematic

Download full size

Fig. 2. RGB+Depth2Mesh model structure schematic

Download full size

Fig. 3. Base network model structure schematic

Download full size

Fig. 4. Dual feature parallel processing algorithm schematic

Download full size

Fig. 5. Multi-scale feature fusion model

Download full size

Fig. 6. Mesh construction module diagram

Download full size

Fig. 7. Depth camera model schematic

Download full size

Fig. 8. Sample grid structured light dataset diagram. (a)‒(c) Projection imaging; (d)‒(f) noise projection imaging; (g)‒(i) depth image

Download full size

Fig. 9. Sample dataset diagram. (a)‒(c) RGB image; (d)‒(f) depth image; (g)‒(i) 3D model

Download full size

Fig. 10. Diagram of qualitative analysis of ablation experiment. (a) Input images; (b) base model output; (c) improved model 1; (d) improved model 2

Download full size

Fig. 11. Model training loss curve

Download full size

Fig. 12. Validation set effect test diagram

Download full size

Fig. 13. Depth prediction effect diagram. (a) Obstacle grid-based structured light imaging diagram; (b) labeled depth map; (c) predicted depth map

Download full size

Fig. 14. Obstacle reconstruction effect diagram. (a) Obstacle type; (b) 3D ground truth; (c) reconstructed 3D model

Download full size

Fig. 15. Actual vehicle installation of structured light sensor

Download full size

Fig. 16. Experimental scene setup

Download full size

Fig. 17. Actual vehicle obstacle 3D reconstruction effect diagram. (a) Traffic cone; (b)(c) traffic cone reconstruction effect; (d) warning board; (e)(f) warning board reconstruction effect

Download full size

Table 1. Quantitative analysis results of ablation experiment

View table

Table 1. Quantitative analysis results of ablation experiment

Model	Evaluation index	Quantized value
Base Model	$F (τ)$	59.735
	$F (2 τ)$	74.216
	$d_{C D}$ /m	0.595
	$d_{E M D}$ /m	1.393
Base Model+U-Net	$F (τ)$	60.320
	$F (2 τ)$	75.401
	$d_{C D}$ /m	0.599
	$d_{E M D}$ /m	1.302
Base Model+U-Net+Double input features	$F (τ)$	61.710
	$F (2 τ)$	76.110
	$d_{C D}$ /m	0.583
	$d_{E M D}$ /m	1.265

Table 2. Transverse comparison of experimental results(CD)

View table

Table 2. Transverse comparison of experimental results(CD)

Category	d_CD/m
Category	3D-R2N2	PSG	N3MR	RGB+Depth2Mesh
Plane	0.895	0.430	0.450	0.441
Bench	1.891	0.629	2.268	0.625
Cabinet	0.735	0.439	2.555	0.350
Car	0.845	0.333	2.298	0.280
Chair	1.432	0.645	2.084	0.613
Monitor	1.707	0.722	3.111	0.731
Lamp	4.009	1.193	3.013	1.218
Speaker	1.507	0.756	3.343	0.748
Firearm	0.993	0.423	2.641	0.451
Couch	1.135	0.549	3.512	0.493
Table	1.116	0.517	2.383	0.501
Cellphone	1.137	0.438	4.366	0.451
Watercraft	1.215	0.633	2.154	0.672
Mean	1.445	0.593	2.629	0.583

Table 3. Transverse comparison of experimental results(EMD)

View table

Table 3. Transverse comparison of experimental results(EMD)

Category	d_EMD/m
Category	3D-R2N2	PSG	N3MR	RGB+Depth2Mesh
Plane	0.606	0.396	7.498	0.401
Bench	1.136	1.113	11.766	0.901
Cabinet	2.520	2.986	17.062	2.511
Car	1.670	1.747	11.641	1.012
Chair	1.466	1.946	11.809	1.108
Monitor	1.667	1.891	14.097	1.290
Lamp	1.424	1.222	14.741	1.501
Speaker	2.732	3.490	16.720	2.801
Firearm	0.688	0.397	11.889	0.380
Couch	2.114	2.207	14.876	1.652
Table	1.641	2.121	12.842	1.380
Cellphone	0.912	1.019	17.649	0.711
Watercraft	0.935	0.945	11.425	0.801
Mean	1.501	1.653	13.386	1.265

Table 4. Quantitative values of depth prediction model
View table
Table 4. Quantitative values of depth prediction model
Evaluation metric Quantitative value
SSIM 0.71
PSNR 30.06

Table 5. Quantitative values of self-built data set model
View table
Table 5. Quantitative values of self-built data set model
Evaluation metric Quantitative value
$F (τ)$ 72.338
$F (2 τ)$ 88.249
$d_{C D}$ /m 0.0069
$d_{E M D}$ /m 0.0138

Table 6. Actual vehicle verification result
View table
Table 6. Actual vehicle verification result
Obstacle Length /m Width /m Height /m
Physical traffic cone 0.40 0.40 0.70
Reconstruction traffic cone 0.31 0.31 0.57
Physical warning board 0.44 0.58 0.62
Reconstruction warning board 0.38 0.53 0.53

Tools

Get Citation

Copy Citation Text

Shidian Ma, Yuxuan Huang, Haobin Jiang, Aoxue Li, Mu Han, Chenxu Li. Three-Dimensional Reconstruction Methods for Obstacles in Complex Parking Scenarios[J]. Laser & Optoelectronics Progress, 2024, 61(24): 2415008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites