UCalib: Cameras Autocalibration on Coastal Video Monitoring Systems

Round 1

Reviewer 1 Report

This paper addresses the problem of camera calibration in coastal monitoring environments over an extended period of time by introducing a methodology for auto-calibration of cameras.  This is then validated through experiments on real world coastal images taken over several years.

The paper is well written and thoughtfully considers common concerns in analyzing the approach.  The results are thorough and do compare to a manual calibration approach as well as provide a thorough discussion.  I only have two minor suggestions.  First, there are a wide range of parameters used in the paper; it could help the reader to have a reference table to refer back to when trying to remember units and variables in reading through.

Second, some of the citations are from the authors.  While it appears that this work builds on and is sufficiently different from prior work, it is appreciated when the paper states that for self-refencing citations.

Author Response

Please see the attached pdf

Author Response File: Author Response.pdf

Reviewer 2 Report

The submitted manuscript describes different steps of an innovative method for automatically georeferencing images captured from three different cameras. I find the work very interesting, well written and relevant to the audience of ‘Remote Sensing Journal. My recommendation is major revision as I think the manuscript in the present form is not ‘mature’ enough for publication even if the authors present their efforts in a good way. I elaborate on this hereinafter.

My main concern is that the authors should address the methodology in a format that could be slightly simplified and shortened. I understand that the reader needs to have all the elements in order to reproduce this method but I think the reader could be sometimes confuse.

• The first things authors could better mention are the different steps of the method at the very beginning of section 2 Methodology. I think that a first short paragraph, coming before section 2.1 “Camera equations and manual calibration”, could really help the reader (who may be inexperienced with imaging-related tools) to follow the entire process. To make it even clearer, the author could add a sentence at the beginning of each section to remind readers the stage we are at. Maybe a synthetic diagram synthetizing all the different steps could also help the reader to understand the method in a more simple way.
• The authors could try to make each section and subsection shorter in methodology (when possible) so the text is more concise and easier to follow.
• A suggestion concerns case 2 in methodology (l. 145-146) referring to a situation that the authors finally do not take into account in their results: ”… the single hypothesis supporting the approach is that the camera position can be regarded to be nearly constant.” (l. 483-484), which is ok to me. Assuming in a more simple way this hypothesis (just one or two sentences) in the methodology could perhaps slightly simplify the content of the manuscript. I leave the authors deciding what seems most relevant to them.
• The author should add a short sentence explaining the concept of “Homographies” in section 2.2.

Another aspect regards the discussion as I think the authors could take a step back from their study and discuss more the applicability of their method to other sites or other types of camera. What are the problems that can be encountered, the advantages and disadvantages of this method comparing with others? Is it applicable everywhere? What are the limitations? I found the discussion a bit long and lacking in perspective while the proposed method and the results are very interesting.

One important thing for a new method I think is the perspective of sharing it to the scientific community. The title of the paper “UCalib: Cameras autocalibration on coastal video monitoring systems” and in particularly the name given to the method “Ucalib” suggest the apparition of a new toolbox able to bring a new solution regarding camera movements and calibrations. I think this paper could largely gain in visibility if their methodology could be shared for instance with the Coastal Imaging Research Network (CIRN) who have deployed a github platform (https://github.com/Coastal-Imaging-Research-Network). If the authors feel the method not mature enough, these aspects should be discussed at least and maybe the title of the paper could be more site specific and less generic.

 

With 20 pages, I find the manuscript in general too long. After simplifying section 2. Methodology, I think the manuscript could even be more precise and pleasant to read by shortening section 3. and section 4. With 9 tables, more than 30 different variables, 6 equations, the manuscript is sometimes very hard to follow. I suggest that the authors try to simplify, for instance by removing or replacing some tables with figures.

---------------

Some minor comments, issues to discuss:

l.1 : “Following the path set out by the “Argus” project”. This directly refers with the work done with the Argus project to initiate larger international collaboration for coastal engineer and researcher with the development for instance of a github platform (https://github.com/Coastal-Imaging-Research-Network). I think this should be mentioned later in the text.

l. 3-5 : “The cameras are always calibrated when installed and, at best, extrinsic calibrations are performed from time to time”. As mentioned later in the text, for few stations an automatic or semi-automatic images calibration method have been developed. Maybe change “always” to “usually”.

l. 10-12 : “The proposed methodology allows for the automatic calibration of > 90% of the images in favourable conditions (images with many rigid features) and ~40% in the worst conditioned camera (almost featureless images)”. Here we get the impression that the authors propose a new method that has been tested in many different environments. However, the method has been only used on three different types of cameras and angles of view, which is ok to me. However, the authors should mention in the abstract where the method has been tested and for how many cameras.

l. 14-16 : “The proposed methodology, here applied to Argus-like stations, is applicable in coastsnap sites, where each image corresponds to a different camera”. Here the authors only mention “coastsnap” but I think the method could be apply on a larger number of tools. Maybe the authors could see further. In addition the non-expert reader does not know yet what is “coastsnap”.

l. 27 : “Qualitative”. Why not “Quantitative”. Please reconsider.

l. 29 : “The images”. Too much “the” in the text. It tends to make the text heavier.

l. 30 : “morfology”. Please reconsider.

l. 45-48 : “Generally, in Argus-like video monitoring systems, a one-time calibration of the system is performed and therefore it is assumed that the calibrations (intrinsic and extrinsic) remain invariant over time”. Not totally agree, in Argus tools, the calibration gui allows the user to create a geometry of the image according to a particular date. This shows that it is not assumed that (intrinsic and extrinsic) parameters remain invariant over time.

l. 55-59 : “To overcome these problems, some camera stations perform periodic extrinsic calibrations in an attempt to correct for long-term variations [11], but without compensating the fluctuations that are constantly occurring“. Not totally agree with authors. Periodic extrinsic calibrations can also be done for each image that will be used for coastal monitoring and this is method still widely practiced. However, as is it is very painful to realize, the frequency of usable calibrated images strongly decrease depending of the capacity of the team using the camera.

l. 91-101 : “Since the points in the real world corresponding to arbitrary pixels are located at unknown heights, it will not be possible to calibrate the images on the standard GCP approach. Alternatively, we relate pixels of pairs of images through homographies and the main assumption of this work is that the position of the camera is time nearly invariant. As a counterpoint, there is no need to impose any constraint on either the intrinsic calibration parameters of the camera (lens distortion, pixel size and decentering) or on the rotation of the cameras. The automatic camera calibration is applied to three video monitoring stations. Two of them operate on the beaches of the city of Barcelona (Spain), where there are many fixed and permanent features, and the third one on the beach of Castelldefels, located southwest of Barcelona, where the number of fixed points is very limited”. This is too long and refer to the methodology. I think this should be removed from the introduction.

l. 109-264 : “2. Methodology”. Please consider my comment above at the beginning of the report.

l. 303-318 : not clear to me. Maybe this will be clearer for me after simplifying the method.

Author Response

Please see the attached pdf

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Editor and Authors, the paper is of interest of Remote Sensing readers. It is well written and structured. Figures are optimally worked. The proposed technique is of great interest and deserves to be published. 

I leave here below some requirements (R) and suggestions (S) to Authors.

R1. You mentioned "coastsnaps" in Abstract, 147 and Conclusions. If you meant the use of smartphones for shoreline mapping, it is missing the reference 10.1016/j.coastaleng.2019.04.003 (and perhaps app and/or repository links), and comments that relate your study to application, at least in Discussion.

Introduction is pleasant to read, however can be improved in terms of reference and details. 

S1 22-23 "monitoring beach nourishment performed
to mitigate coastal erosion" can be https://www.mdpi.com/2073-4441/12/6/1632 

S2 24-27 you mentioned "early 80's" but report only references from 2000's. I suggest to add here also [21], [38] ecc. 

S3 [10-15] references are relevant but dated, I believe (many) more recent works and applications can be added, since video monitoring has been improved in the last 15 years.  

R2 44-45 "The GCPs can also be used for intrinsic calibration,
which is however usually obtained experimentally in the laboratory [20]." The term "usually" is perhaps too strong here, citing a 1999 reference. You properly mentioned in the next paragraph the one-time calibration in the field, however I believe it is missing to report the C-pro  10.1016/j.isprsjprs.2017.03.023 and its application in the field doi.org/10.3390/rs11010078 , for instance, which calibrated the camera through horizon and GCPs in the field.

S4 Methods are well explained. Perhaps some paragraphs can be deleted or shorten, especially when you introduce the sections (e.g. 104-108, 155,160,214-222). I leave it as suggestion. 

R3 It would be of interest to see/read, at least in Discussion, what is (if there is) the relation among fC and Kc  (or fC = 5 pixel and KC= = 4 proposed for auto-calibration), the image size and camera properties. Image sizes (different for each station) are given just in Fig.4 caption (I suggest to mention it also in the text), however for future applications it would be of interest to clarify/comment that. From Table 6-9, it is clear that there are differences in the results, but it is not clarified in the text the role played by image size and/or camera movements in the difference stations. 

R4 Fig.11 is much relevant (please, add colorbars labels), and its description should be improved. Is is not clear to me from Fig., the real impact of improved calibration.

S5 It would be interesting to see the real impact of auto-calibration, for instance, on a shoreline marked on a generated image plan. Since spatial resolution on image plan rapidly decreases with distance, I wonder to what extent calibration/non-calibration would influence the marked shoreline (or bathymetry retrieval as mentioned in the paper) at 500 m, for instance, where the spatial resolution is already low. This cannot be really seen in Fig.11.

 

Author Response

Please see the attached pdf

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Dear Editor and Authors, the new version of the paper have been improved following my recommendations and suggestions. The paper can be published.