Three-Dimensional Reconstruction of Ocean Waves Based on Mask and Self-Supervised Learning

Junjie Huang; Feng Xu; Liang Luo; Tianbao Chen

doi:10.3788/LOP231953

Laser & Optoelectronics Progress, Volume. 61, Issue 14, 1437008(2024)

Three-Dimensional Reconstruction of Ocean Waves Based on Mask and Self-Supervised Learning

Junjie Huang^1,2, Feng Xu^1,2、*, Liang Luo^1,2, and Tianbao Chen^1,2

Author Affiliations

¹School of Information Engineering, Southwest University of Science and Technology, Mianyang 621010, Sichuan , China

²Robot Technology Used for Special Environment Key Laboratory of Sichuan Province, Mianyang 621010, Sichuan , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Network structure

Download full size

Fig. 2. Schematic of loss calculation

Download full size

Fig. 3. Common area disparity mask. (a) Left mask; (b) right mask

Download full size

Fig. 4. Results of disparity estimation. (a) Input image; (b) WASS; (c) Monodepth-LR; (d) Monodepth-2; (e) proposed method 1; (f) proposed method 2

Download full size

Fig. 5. Comparison of mask shielding effects. (a)(b) Input images; (c)(d) WASS; (e)(f) Monodepth-2; (g) (h) proposed method 1; (i) (j) proposed method 2

Download full size

Fig. 6. Comparison of point cloud generation effects. (a) WASS; (b) Monodepth-LR; (c) Monodepth-2; (d) proposed method 1; (e) proposed method 2

Download full size

Fig. 7. Comparison of grid mapping effects. (a) WASS; (b) WASSfast-grid; (c) WASSfast-fft; (d) proposed method 1; (e) proposed method 2

Download full size

Fig. 8. Sea surface elevation and deviation distribution. (a) (b) WASSfast-grid; (c) (d) WASSfast-fft; (e) (f) proposed method 1; (g) (h) proposed method 2

Download full size

Table 1. Model complexity and speed evaluation
View table
Table 1. Model complexity and speed evaluation
Method Image size FLOPs /10⁹ Params /10⁶ Speed /ms
Monodepth-LR 512×448 13.88 20.22 16.83
Monodepth-2 512×448 21.30 17.47 20.60
Proposed method 1 512×448 10.70 3.29 18.24
Proposed method 2 1216×1024 14.67 3.30 24.14

Table 2. Sea surface elevation statistics

View table

Table 2. Sea surface elevation statistics

Method	Z<1 m	Z <1.25 m	Z <1.5 m
WASS	97.3725	99.5437	99.9622
WASSfast	98.3899	99.8064	99.8268
Monodepth-LR	83.0713	84.4159	84.9295
Monodepth-2	84.2286	86.6222	87.2233
Proposed method 1	98.8942	99.7630	99.9985
Proposed method 1	99.2194	99.9423	99.9947

Table 3. Statistics of different elevation sequences

View table

Table 3. Statistics of different elevation sequences

Method	Correlation	Mean error /m	Standard deviation error /m	Max error /m
WASSfast-grid	0.9534	0.0885	0.0734	0.6600
WASSfast-fft	0.9620	0.0783	0.0661	0.5356
Proposed method 1	0.9747	0.0779	0.0614	0.4877
Proposed method 1	0.9775	0.0624	0.0489	0.4518

Table 4. Comparison of time consumption between different algorithms

View table

Table 4. Comparison of time consumption between different algorithms

Method	Image size	Point cloud	Reconstruction /s	Grid /s	Total /s
WASS	512×448	179789	0.295	1.219	1.514
WASS	1216×1024	986488	6.395	6.186	12.581
WASS	2468×2056	3194012	27.202	25.139	52.341
WASSfast-grid	1200×1200	10380	1.007	0.044	1.051
WASSfast-fft	1200×1200	10380	1.007	1.568	2.575
Proposed method 1	512×448	182517	0.024	1.036	1.060
Proposed method 2	1216×1024	939745	0.059	5.748	5.807

Tools

Get Citation

Copy Citation Text

Junjie Huang, Feng Xu, Liang Luo, Tianbao Chen. Three-Dimensional Reconstruction of Ocean Waves Based on Mask and Self-Supervised Learning[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1437008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites