ROV Navigation in a Fish Cage with Laser-Camera Triangulation

Round 1

Reviewer 1 Report

The manuscript (jmse-1050225) presents a novelty positioning system based in laser-camera triangulation, in this case applied to aquaculture net inspections using autonomous underwater vehicles. It is an interesting topic with many possible applications, since underwater positioning is one of the main problems for the underwater vehicles to navigate autonomously.

The method is well explained and the calibration process is clear. I have some doubts. How is affected the camera calibration by the distance between the camera and the object, in this case, the net. How affects the refraction in this case? Also, it is suggested to show the calibration parameters obtained. For readers like me, that works with photogrammetry but not underwater photogrammetry, so more information about the difference between underwater calibration and in-air calibration will be appreciated.

The following paper could help you: https://0-doi-org.brum.beds.ac.uk/10.1007/978-3-030-03635-5_2

The system is tested in a real study case, navigating inside of a fish cage. Why the system navigates inside instead of navigating outside the net? In this way the noise due to the fish can be decreased or avoided.

Conclusion section should be extended, adding more numeric data.

Author Response

Thank you for your review, we have expanded the paper to answer three of your questions:

1) The camera calibration method is covered in more detail, and we have commented the errors introduced by the underwater setup and how we handle those. (section 2.4)

2) We explain that the inside-net ROV operations are due to obstructing ropes/moorings around the net. (section 2.3)

3) A comparative noise estimate is added to have more numeric data in the conclusion. (conclusion, and section 3.4)

best regards

Magnus Bjerkeng

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper presents an approach to estimate the pose of an AUV or a ROV with respect to an underwater fish cage by means of laser-camera triangulation techniques. These scenarios are problematic due to their changing shape and visual appearance with time.

Overall the paper is well written and structured, it discusses an interesting problem perfectly suitable for this journal and special issue, and it proposes a working solution. The proposal is tested experimentally in a real scenario.

My main concerns are twofold. On the one hand, some parts of the background should be expanded to provide the reader with a clear view of the area. On the other hand, some parts of the specific research contents should be explained with more detail. More specifically:

Background
==========

The authors state that there are not industrial deployments of autonomous net inspection and, so, they survey some sensors that are general purpose. Among them, they talk about structured light systems, including existing laser triangulation systems. Even though the information seems correct, I'd ask the authors to improve and expand the review on existing structured light systems and, more specifically, laser triangulation systems, focusing on their advantages and disadvantages with respect to the authors proposal.

The paper includes a link to one commercial laser triangulation system (by 2G Robotics) and they claim that this sensor requires mechanical scanning, thus having a large scan time. I'd also ask the authors to be more specific: why mechanical scanning is required in this case and not in the paper proposal? is the light pattern of these devices different to the one used in the paper?

Fish cage inspections is a very specific task and, so, some aspects should be explained with more detail. For example, the cage walls are always squared grids? How large are the holes? Typical inspection tasks are performed with the AUV inside the cage (as in the experiments) or outside? Why? Are usual cages double net or single net? That's particularly important, since performance (according to the provided results) is affected by the double net.

Also related the previous point: how is the cage inspection done? does a human operator searches the video stream for problems in the net? are defects detected automatically by means of computer vision?

Even though the advantages of the authors proposal with respect to DVL are more than clear, it is unclear that a structured light system like this is the best choice in these environments (nets). So, please, clarify why a structured light system has been selected instead of (among others) stereo vision, non-DVL acoustic sensors (such as sonar scanners or even pingers), ...

Research contents
=================

Please, clarify from the beginning that approximating the cage walls by planes is only a local approximation (not a global one). Otherwise, it seems an extremely unrealistic assumption.

Why have you chosen two parallel lines instead of (for example) perpendicular lines or some other, more complex, pattern? Please, provide some information.

The only information provided about how the planes are extracted from the lines is a reference to MLESAC. Some more discussion is expected in this regard: do dots in the two lines have to be associated? if so, how is this association done? how to the grid holes affect the process? how does the algorithm work? A very basic description about the algorithm would suffice, but should be provided anyway.

Some more information about the specific net used in the experiment should be provided: how large are the holes? is the net material dirty or clean? dark or clear? is it homogeneous, or there are clearly differentiated patches? were there water turbulences during the experiment? fish population was normal? is there a double net along the whole cage or only in certain parts?

The provided results are sufficient to see how the authors proposal behaves. The discussion, however, is too qualitative. I understand that an error measure is not possible (since there is no ground truth) but I'd like to see more quantitative information: maybe some confidence intervals to determine the percentage of outliers, maybe a measure of the data dispersion, ... I'd just like to have some statistics comparing the authors proposal and the DVL.

Finally, it would have been nice to have some CPU consumption results (time consumption, among others). I am not going to ask these results, since they are not really important given the nature of the paper, but I'd like to see some discussion: what is the approximate processing time per frame? on what CPU? does the on-board computer suffice? or did you have to replace it by another one? have the experiments been performed by the on-board computer?

Small errors/typos
==================

There are a few typos/minor errors that should be corrected:

* Page 1: please remove the comma before the [2] at the end of the first paragraph.
* Page 2, line 9: "Flash-lidars does not include" -> "Flash-lidars do not include.
* Please, use LIDAR (capital letters) instead of lidar.
* Page 3: SINTEF is used within the text here for the first time. Please, state here what is SINTEF (something like "our company" or "the company SINTEF" or something like that)
* Page 4, first paragraph: please explain (or expand the acronym) "DP" (in the ROV was placed in DP mode).

As a final remark, I'd like to emphasize that this is a good paper. None of the remarks I've made is related to the work but to the paper itself, which may require only some minor changes.

Author Response

Thank you for your thorough and fair review, we have edited the manuscript as follows:

The introduction now explains why we do not need a scanning sensor, (it is more expensive, and not needed for detecting planes).

The introduction as well as section 2.3 explains why inspections are carried out inside the net and not outside. (because of obstructing infrastructure). More details on the fish nets are also given. The following excerpt contains many of the additions: 

"ROV operations in fish farms are commonly performed inside the fish cages, not outside. This is due to the presence of ropes, chains and mooring lines on the outside of the cages which the ROV’s tether can get tangled into. When operating inside the cage, the fish will sometimes obstruct the cameras and sensor measurements, but the severity of this compared to the ROV being tangled, is low. The cage has a cylindrical shape with a conical bottom, where the upper diameter is approximately 50 m and the total depth is about 30 m. The cage used in the trial is equipped with double nets in the regions around the main ropes to secure these regions against fish escapes. This double net setup is used in all fish cages operated by the company operating at SINTEF ACE, but it is unknown if this is standard for other companies. As will be shown, the presence of the double nets influences the quality of the distance measurements when using the DVL. The nets at the Tristeinen facility are square, with a mesh width of 33 mm. They had been cleaned 8 days prior to the trials. The nets originally had a green coating, but some of this seems to have worn off. Fish population was approximately 190.000 individuals, which is normal."

Section 2.4 contains more details on the laser triangulation /structured light setup.

The introduction now also gives more details on how the net ispection is done today: 

"Net inspections today are commonly performed either by divers or by human-piloted remotely operated vehicles (ROVs). The ROV operations are challenging for the pilot as they require both precise maneuvering and a keen eye for detail in order to detect failures in the net cage from the video stream."

The introduction was expanded to explain why we do not use stereo-cameras, which would be even cheaper than the camera laser setup.  
A more expansive comparison with other acoustic sensors was not done.

Finally your 5 syntax errors have been corrected. 

best regards 
Magnus Bjerkeng

Author Response File: Author Response.pdf