Optical Guidance for AUV Underwater Docking (Invited)

Fig. 5. Platform and cage docks. (a) Platform dock; (b) contour extraction of platform dock; (c) marker detection and recognition of platform dock; (d) dock with light beacons combined with triangular black and white codes

Download full size

Fig. 6. Identification of underwater guidance markers and light beacons. (a) Classification and identification of markers; (b) identification of 8 lights

Download full size

Fig. 7. Underwater binocular vision positioning principle. (a) Pixel coordinates of camera at guide light center in dock; (b) principle of binocular stereo vision

Download full size

Fig. 8. Research on binocular vision guidance technology by Myint’s team^[111-115]. (a) ROV binocular vision guidance pool test; (b) ROV binocular vision guidance sea trial; (c) active light-emitting 3D beacon; (d)‒(f) 3D beacon images and recognition results captured by binocular cameras in water with different turbidity

Download full size

Fig. 9. Distance-exposure time calibration chart and effectiveness comparison between fixed exposure and automatic exposure. (a) Comparison between underwater unpolarized imaging and polarized imaging; (b) comparison of effects between underwater fixed exposure and automatic exposure

Download full size

Fig. 10. AUV underwater recovery positioning and optical guidance tests. (a) Onshore test of guidance lights; (b) recognition of guidance lights by binocular cameras; (c) guidance light structure design; (d) AUV guidance head design diagram; (e) optical guidance positioning bracket; (f) onshore attitude and azimuth test; (g) underwater attitude and azimuth test; (h)(i) underwater AUV docking tests

Download full size

Fig. 11. Principle of four-quadrant photodetector

Download full size

Fig. 12. Optical guidance detector elements based on photoelectric detection arrays. (a) REMUS 600 with four-quadrant photodetectors; (b) hemispherical 5×5 optical detector array; (c) 7 photodetectors and lenses form a hemispherical array

Download full size

Fig. 13. Diagrams of AUV guidance location system. (a) AUV underwater recovery optical guidance process; (b) AUV coordinate system diagram; (c) guidance light deflection angle geometric definition; (d) guidance light coordinate system diagram; (e) ROV remote arming; (f) real-time solution and location; (g) daytime sea trial guidance location; (h) night sea trial guidance location

Download full size

Table 1. Summary of status of monocular visual guidance and localization methods

View table

Table 1. Summary of status of monocular visual guidance and localization methods

Author	Experiment environment	Range	Frequency or time	Evaluating indicator	Computing platform	Ref. No
Park et al.	Pool	≤10 m	10‒15 Hz	Average position error incamera coordinate: 200 pixel	PC/104+ standard single-board computer	[11‒12]
Palomeras et al.	Tank	≤10 m	15 Hz	—	—	[13]
Fan et al.	Pool	≤2.8 m	—	Lateral positioning error is ≤0.2 m; longitudinal and vertical positioning errors are ≤ 0.15 m	—	[14]
Zhang et al.	Pool	—	—	Position error is 0.1 m	—	[16]
Trslic et al.	Sea	—	—	—	PC	[17]
Figueiredo et al.	Pool	≤2.5 m	20 Hz	Position error is 0.1 m	Raspberry Pi	[19]
Zhang et al.	Pool	≤10 m	30 s	—	—	[23]
Zhang et al.	Pool	≤5 m	50 s	—	—	[24]
Rannnestad et al.	Sea	≤6 m	26‒30 Hz	—	—	[20]
Lv et al.	Pool	≤10 m	—	Position error (X or Y) is ≤80 mm Position error (Z) ≤ is 140 mm	—	[21]
Yan et al.	Pool	—	3.07 s	Matching accuracy is 100%	Intel^® TM6600 CPU, 2.4 GHz, 1.0 GB EMS memory,	[22]
Zhang et al.	Pool	≤1.3 m	13.5709 ms	—	ThinkPad P52s machine	[23]
Lin et al.	pool	≤1.5 m	35 Hz	Error range is 3‒20 cm	PC	[26]
Zhang et al.	Lake	≤6 m	—	Angle error is 1°	—	[27]
Zhou et al.	Tank	—	—	—	—	[40]
Wat et al.	Air	≤7 m	30 s	Average angle error is 10°	PC	[41]

Table 2. Summary of status of deep learning based monocular visual gudiance and localization methods

View table

Table 2. Summary of status of deep learning based monocular visual gudiance and localization methods

Author	Method	Environment	Range	Frequency	Evaluating indicator	Computing platform	Ref. No
Singh et al.	YOLO	Pool	≤15 m	6 Hz	Average location error is 890 mm	NVIDIA Jetson TX2	[28]
Ren et al.	YOLOV3 & P4P	Pool	≤10 m	—	Average location error is 30 mm	—	[30]
Zhang et al.	SNN & P4P	Pool	—	—	—	—	[33]
Jyothi et al.	YOLOV5	Tank	≤10 m	—	Success rate is 95%	128 core NVIDIA Maxwell edgeArm cortex-A57 MP	[36]
Lu et al.	YOLOV5	Pool	≤15 m	—	—	Intel^® Xeon^® Silver 4110 CPU @2.10 GHz 16 GB GeForce RTX 3060 Ti 11 GB	[37]
Liu et al.	CNN	Sea	—	—	Success rate is 88%	—	[40]
Yahya et al.	CNN	Tank	≤8 m	60 Hz	Success rate is 95%	2.6 GHz Intel Core i7	[42]

Table 3. Summary of status of camera calibration methods

View table

Table 3. Summary of status of camera calibration methods

Author	Complexity	Robust	Cost	Accuracy/error	Ref. No
Ge et al.	Low	Low	Low	0.19 pixel	[55]
Chen et al.	Middle	High	Middle	0.05 pixel	[56]
Zhang et al.	Low	Middle	Low	1.3 pixel	[57]
Zhang et al.	High	Middle	High	0.03 cm	[58]
Hartley	Low	High	Low	0.5 pixel	[59]
Yang et al.	Low	High	Low	Low	[60]
Wu et al.	Middle	Middle	Middle	Low	[61]
Wu et al.	Low	Middle	Low	Low	[62]

Table 4. Summary of status of binocular vision guidance and localization methods

View table

Table 4. Summary of status of binocular vision guidance and localization methods

Author	Environment	Range	Frequency/time	Evaluating indicator	Computing platform	Ref. No
Myint et al.	Pool	(0.35 m,2 m)	165 ms	Average location error: 40 mm	ROV	[111]
Myint et al.	Sea trial 50 m depth	(0.35 m, 15 m)	33 ms	Average location error: 10 mm	ROV	[115]
Shiet al.	Lab	—	258 ms	Average location error: 5 m	Pentium 42 GHz	[129]
Liet al.	Pool	≤30 m	0.5 s	Average location error: 300 mm	WL-3 mini-micro AUV	[130]
Zhuet al.	Pool	≤10 m	7.5 ms	Average location error (@10 m): 140 mmAverage location error (@5 m): 92 mmAverage location error (@3 m): 53 mm	—	[132]
Xuet al.	Pool & lake	Pool: ≤20 mLake: ≤7 m	21 ms	Average location error (X, pool): 14 mmAverage location error (Y, pool): 7.8 mmAverage location error (X, lake): 2.8 mmAverage location error (Y, lake): 3.2 mm	Intel i7-8750 processor and an NVIDIA GTX 1060 graphics	[133‒134]
Guoet al.	Pool	≤2.81 m	—	Average location error: 200 mm	Amphibious spherical robot	[138]
Hsuet al.	Tank	≤0.95 m	—	—	—	[139]

Table 5. Summary of status of optically guided localization methods based on position detectors

View table

Table 5. Summary of status of optically guided localization methods based on position detectors

Author	Environment	Range	Frequency/time	Evaluating indicator	Computing platform	Ref. No
Martin et al.	Sea trial	≤10 m	—	Recovery success rate: 48.40%	—	[152]
Fletcher et al.	Sea trial	≤10 m	—	Recovery success rate: 41.70%	i7‒3770 CPU	[153]
Evans et al.	Pool	≤5 m	1 Hz	Positioning accuracy: 0.2 m; detection angle: 10°	—	[156]
Erenet al.	Pool	≤8.5 m	5 Hz	Positioning accuracy of x, y and z axes: 0.13 m, 0.13 m and 0.18 m; pitch and yaw errors: <5°	—	[157]
Ours	Simulation & Sea trial	0.8- 20 m	5.560 ms	0.8‒20 m absolute coordinate error: 58.292 mm; ≤3 m absolute coordinate error: 6.647 mm	ARM Cortex‒A78AE v8.2 64 CPU	[160]