Photogrammetry - Science topic

Thanks for responding Fulgence Hatangimana your answer provides a brief overview of first doing the radiometric correction by following the the GitHub repo by Micasense. However, my question was trying to get into the details of radiometric correction that is done by Metashape. For example, if I apply the row gradient correction before processing my images in Metashape, how do I know that Metashape is not trying to repeat the same correction?

Aahed Alhamamy , the author Dr Abdullah, had this to say: "The new equation in edition 2 differs from the one we published in edition 1 as it is based on two accuracies for the IMU. In the earlier version, the equation assumes one accuracy for the roll, pitch, and heading. Since the accuracy of the heading is always worse than the ones for roll and pitch, we revised the equation, so the user can enter two values for the IMU accuracy."

Yes, without GCPs and RTK/PPK, it is highly possible to obtain wrong or inaccurate geometric information in UAV-based photogrammetric mapping. When dealing with large areas, relying solely on the drone's GNSS information can lead to distortions in the 3D model or orthomosaic. GCPs not only add global georeferenced accuracy but also help decrease errors in the scale of the results, making them essential for reliable measurements.

The accuracy of UAS-based photogrammetric mapping depends on several factors, including Ground Control Points (GCPs), flight height, camera resolution, GNSS accuracy of the device, weather conditions, processing software, user experience, etc. While it is possible to obtain satisfactory 3D models without GCPs or RTK/PPK, for precise measurements such as surface area or volume calculations, GCPs are essential. For more in-depth information, you can refer to my MSc thesis and the following articles. In the following studies, you can find comparisons of processing with and without GCPs as well.

MSc Thesis:

Good luck on your journey of exploration and innovation!

Here are some best practices to follow when using these cameras for 3D photogrammetry:

Ziheng Zhang In the engineering profession, there are several possible uses for extended reality (XR). Among the options are:

- Engineers can utilize virtual reality (VR) to construct and test virtual prototypes of items before building them in the physical world. This can save engineers time and money by allowing them to detect and correct design flaws before manufacturing physical prototypes.

- Training and education: Engineers and other technical workers can benefit from immersive training simulations created with virtual reality (VR) and augmented reality (AR). This may make learning complicated topics and procedures more entertaining and successful.

- Remote collaboration: XR may be used to connect engineers in various places so that they can collaborate in a shared virtual environment. This can help with project communication and cooperation regardless of location.

- Maintenance and repair: Using AR, engineers may get real-time information and directions while doing maintenance or repairs on complicated equipment.

- Site inspections: With XR, engineers may conduct virtual site inspections and discover possible concerns before they exist.

Overall, the application of XR in engineering can assist to simplify processes, enhance efficiency, and promote field accuracy and safety.

Helpful to your analysis would be LiDAR coverage taken during low flow to obtain some channel morphology and as well as topographic detail. Dr. David Rosgen has developed a stream classification system that uses features as gradient, entrenchment, bankfull elevation (top of point bars that might be visible with LiDAR), braided vs non-braided channels, sinuosity, etc. The infrared aerial photos are often useful in separating upland and bottomland species if forested. The LiDAR very helpful in obtaining channel detail that is obscured by dense forest vegetation. Soil scientists familiar with mapping soil types on landscapes in your area would be helpful as they are used to reading landform, vegetation and hydrology indicators. But they too need experience and ground truthing (checking, transects, site visits, etc.).

As suggested by Dr. Klebs, securing available rainfall and stream/river data for your physiographic area(s), developing detailed DEM, classifying vegetation and channel types, Watershed sizes for areas of interest, land use practices, land cover types can also be helpful. Braided and highly sinuous streams almost always have significant flooding. Highly entrenched and high gradient streams tend to have no or narrow floodplain. If a flat area is adjacent to a stream or river, it could be floodplain, unless the channel is deeply entrenched, then it is likely a terrace (abandoned floodplain). Braided streams are typically a result of excess sediment loading, which fills channel capacity and aggrades channel, resulting in more frequent flooding and channel instability.

Due to the potential for land valuation and regulation, as well as human hazards, it may be best to have licensed engineer and appropriate agency review and approval before publishing results that potentially could have legal or human ramifications. If not, perhaps publishing as guideline or remote estimates needing field checking and verification would help limit liability for misinterpretation. Without LiDAR, there are certainly things one can obtain from aerial photos and photo pair interpretation, such as landform, vegetation type and some channel indicators perhaps. Soil maps if available typically separate out flood prone and hydric soils that would be useful information.

Therefore, you have 20x20 = 400 control points. If you do georeferencing in Qgis, you can use all control points or some of them, like every 5 Km (16-points). During resampling, all pixels have coordinates in the ground system.

If you do not do georeferencing (no resampling), then you could calculate the coordinates of unknown pixels by interpolation. Suppose a pixel size a [m], then in one km, you have p = 1000/a pixels, and therefore known coordinates have the first(x1,y1) and the last(x2,y2) pixel. The slope angle between the first and last pixel is:

s = arc-tan[(x2-x1)/(y2-y1)]. Therefore, a pixel of a distance d from the first pixel has coordinates x = x1 + d.sin(s) and y = y1 +d.cos(s). You can do either row of column interpolation or both and take the average.

Me neither, I'm sorry by that.

Regards

Are there any videos showing the steps of comparing two stl. models using Meshlab?

At aerial image,

Ground Sampling Distance (GSD) = one pixel side dimension (sensor) X (average camera height / focal length )

GSD = pixel side dimension X (1/ average imaging scale)

heck this paper for illustration

Low Cost Technique of Enhancement Georeferencing for UAV Linear Projects | Proceedings of the 2020 3rd International Conference on Geoinformatics and Data Analysis (acm.org)

Dear Martin Mokros,

Got this Project！ I‘ll share this with corresponding workmates

Thanks for sharing this info. Kind Regards!

If the area is relatively flat, you can use a single photograph with projective transformation being an option in georeferencing using the QGIS. The accuracy depends on the magnitude of elevation differences from a fitted plane to the earth's surface. Such elevation differences create the so-called relief displacement that produces the error. Such error analysis is shown in the Chapter of Photogrammetry in the book Topographic Mapping, Universal Publis, 2008, pp 349.

In RS, you can develop some methods to extract features from satellite images.

Dear Ivan Marić,

Please check my Researchgate and you will find a paper that I published with my students using DJI phantom-4 for 3D mapping for part of the railway station at Khartoum-Sudan. We used the RTK-GPS to provide the ground control points. And we do have a case study in the paper.

Regards,

Gamal

Please take a look at Dotproduct DPI-10. It uses the latest intel sensors and SLAM algorithms.

Hi Amir,

the following papers provide some interesting info on the topic of roughness and how to compute it:

* Kushunapally, R., Razdan, A., Bridges, N., 2007. Roughness as a Shape Measure. Computer-Aided Design and Applications 4 (1-4), 295–310. DOI: 10.1080/16864360.2007.10738550 (download here: http://www.cad-journal.net/files/vol_4/Vol4Nos1-4.html)

* Nocerino, E., Stathopoulou, E.K., Rigon, S., Remondino, F., 2020. Surface Reconstruction Assessment in Photogrammetric Applications. Sensors 20 (20). Doi: 10.3390/s20205863 (Download here: https://www.mdpi.com/1424-8220/20/20/5863)

As these paper state, one first must define if a local or global roughness measure is needed. Recently, machine learning approaches have been leveraged to estimate roughness, but the more common approach is to use 3D-capable tools to compute and visualize roughness. Two good and free tools are:

* CloudCompare, which can directly compute local roughness (info here: http://www.cloudcompare.org/doc/wiki/index.php?title=Roughness, download here: https://www.danielgm.net/cc/)

* Meshlab, like CloudCompare a tool that is capable of many mesh and point cloud operations (download here: https://www.meshlab.net/#download). Although there is no direct “roughness” measure in Meshlab, the software offers several ways to compute surface curvature, which can be used as a proxy for roughness. You will find some useful information here: https://stackoverflow.com/questions/59194552/meshlab-does-the-topology-of-my-mesh-effect-the-curvature-results

Hope that helps. Cheers!

Geert

Here are a few resources that should be helpful:

1. The book Structure from Motion in the Geosciences: https://www.wiley.com/en-us/Structure+from+Motion+in+the+Geosciences-p-9781118895849

2. Free short courses from UNAVCO, Here is one example short course with materials that you can view and download: https://kb.unavco.org/kb/article/gsa-2019-short-course-510-high-resolution-topography-and-3d-imaging-ii-introduction-to-structure-from-motion-sfm-photogrammetry-869.html

3. Typically, UNVACO and OpenTopography have a short course on SfM at the GSA Annual Meeting (See the reference above for thee 2019 GSA Annual Meeting short course). I would check the GSA Annual Meeting website to see if UNAVCO/OpenTopography will be holding another short course this year.

Hope this helps.

Dear Krishna Prosad Mondal

These articles might be helpful, have a look:

1. https://res.mdpi.com/d_attachment/remotesensing/remotesensing-10-01952/article_deploy/remotesensing-10-01952.pdf

2.

Kind Regards

Qamar Ul Islam

Thank you very much. I will check both of them.

Hi Heather,

can you clarify if by photogrammetry for conservation you mean applications for a) documenting heritage objects for dissemination/virtual display/augmented exhibition, or b) contactless study and/or vistual reconstruction, or c) identifying and mapping decay, previous conservation interventions etc.? because these are usually different categories of documentation in terms of quality and product requirements.

+1 to Arnadi Murtiyoso 's comment

regards,

Efstathios

I would say there is significant progress in feature based photogrammetry.

I can think of 2 reasons why it might be thought there is no progress, first, is there a market need for it, most systems are developed in conjunction with industry to meet their need and then could spin off as a product that could serve or be applied to that type of industry. Second, current applications require a knowledge base that spans different fields of study and this might be a major limiting factor to conceptualization, algorithmic formulation and system implementation such that there is its acceptance within the framework of a particular industry, hence development of such systems would occur at mostly MSc-PhD level. A simple example is the development of a mapping system for underground mines. Such systems exist, and require ability to merge knowledge from computer science, robotics, geomatics/surveying, mining and perhaps other fields to be acceptable into a traditional industry that requires accuracies of a certain degree. Data inputs for an underground system will differ drastically from that of a terrestrial mapping system or an air-borne system even though the concept is similar. Again there most likely is no common solution because of the data inputs from the different instruments that can be used to collect data.

Essentially, the need for a system is supported by available technology and the human knowledge base to bring the system into being.

As stated in your question, you need to measure the image coordinates from a diapositive (assuming that this a classical aerial image). First you need to determine the coordinates of the fiducial marks and use their known values to compute the parameters of affine transformation. The end end results of this work is to define the mathematical center of the image. Then during your collinearity computation you need to include the coordinates of the principal point as unknown in order to perform an implicit transformation to the physical center of the perspective geometry that is defined by the coordinates of the principal points. In some cases you may have the coordinates of the principal points known from a previous camera calibration process and in this case you need to impose their values explicitly in a reduction process in order to transform from the fiducial center to the principal point center. On the other hand, if you object space is not planner, you can use the Direct Linear Transformation (DLT) to work directly without any concern about the center of y=our image coordinates since the DLT will perform to transformation simultaneously (Affine + collinearity).

You can take 3D scan images using a mobile camera and a simple software. Today, software has been created that you can not give images of different views, and this software turns it into a three-dimensional object. This is the cheapest form possible. It's actually free

demography or urban extensions

Dear Abdul,

I did a similar research some time ago, where I used a laser scan derived-mesh model as reference with which to compare the results of some photogrammetric software and/or techniques. Alternately, you can also create an average surface mesh from all your photogrammetric point clouds to use as a reference.

I used the Mesh-2-Cloud and MC3D tools in CloudCompare to generate a Gaussian curve of the error between my reference and compared point clouds. From the Gaussian curve you can then infer average error and standard deviation with regards to a common reference, thus deducing the quality of each software.

You should be careful when comparing photogrammetric point cloud density because it would depend on the settings/parameters you used for each software.

Here are some of my papers on the subject:

(last paper is in Bahasa Indonesia).

Hope this helps.

Best,

arnadi

Marshall Greenspan

Thank you for your answer.

Is it necessary to center the images to the Doppler centroid before using the Fourier transform?

For simple geometrical shapes a number of metrics like area, perimeter, eccentricity, bounding box etc. can be easily calculated through image processing techniques to determine shape characteristics. For irregular shapes techniques like centroid-radii model can be used.

@Qayssar Ajaj

It is undeniable that photogrammetry is a very effective non-contact observation method. However, it also has some limitations:

It is undeniable that photogrammetry is a very effective non-contact observation method. However, it also has some limitations:are equipment, such as the image pixel quality of camera;

2. The matching algorithm of stereo image pairs, such as image batch matching algorithm and scale conversion, is difficult;

3. Photogrammetry does not have penetration ability for observation with occlusion, for example, canopy occlusion makes it difficult for photogrammetry to obtain tree factors above and below the canopy at the same time;

4. For terrestrial photogrammetry, the operator's photography skills are required.

5. In addition, photogrammetry also has certain mandatory requirements on the observation environment, such as illumination, rainfall, cloud and fog.

Hopefully you have solved it already but if not try cloudcompare.

Hi Ghasem Askari , will try your suggestion...but using such approach the resulting georreferenced pictures will have a good accuracy only in them central parts. Isn't it?

Dear Kasiviswandham Ponnapalli,

what i recommend first ist to understand between the difference of a 2D and a 3D map. 2D Maps are mainly for the path planning and integration of objects, 3D maps are mainly for the localization of the vehicle. The paper i recommend are the following:

Dear Mirko,

Please a look to the attached paper, which may address your question. It is about using multi-resolution image representation for texture analysis. I used MATLAB to implement the presented idea. But it is not difficult to implement it in ArcGIS. These are so many applications that can benefit from the instanteneous use of multi-resolution image representation such as image matching and the detection of the intrinsic scale of image features.

Many thanks Al Mouayed, you have saved me a lot of time. Always grateful!

Kind Regards,

Hasanain

Jorge, you may be interested in this article by Cook and Dietze (2019)

https://www.earth-surf-dynam-discuss.net/esurf-2019-27/esurf-2019-27.pdf

They propose a system to compare UAV derived point clouds without ground control essentially by using the sparse point cloud from Survey 1 as control for Survey 2. As photogrammetric datasets can be subject to distortion, this essential would give the same distortion to both datasets, so change detection results are more representative of process than reconstruction error.

As mentioned by Edgar, automated or pick-point registration can be a nice approach too, but since you are working with a glacier where surfaces could potentially retreat at a similar rate, you might overfit your data and estimate lower retreats than actually happened. To get around that, you could focus on bounding rocky sections of the valley to get your translation matrix (the geometric matrix used to move your data to the best fit in CloudCompare) and apply the best translation to your entire dataset for change detection.

thanks Aysar Jameel Abdalkadhum

Referencing models with GCP is a great start, and if the GCPs are the exact same positions for both datasets, that is even better. To reiterate what I think your question is, you want to know 1) are GCPs enough for registering two datasets for topographic change detection, and if not, 2) is automated registration a better option for reducing error.

To make this decision, we need to know about GCP coordinate accuracy and the distribution and number of GCP. I will note that when I say coordinate accuracy, I refer to the accuracy of the instrument, such as GPS or total station, and not the accuracy reported within a photogrammetric project. If you have a lot of points (>10) and they are well distributed through your area of interest and your coordinate accuracy is high (~3 cm), then GCP is probably sufficient. If not, it may still be sufficient, you just need to test it with alignment for areas that likely have not have changed (invariant or pseudo-invariant points) or adjust you accepted minimum detectable change threshold - sometimes defined as the propagated error of your GCP (SQRT(error1^2+error2^2).

This error can be improved with automated registration, like the cloudcompare fine registration tool, but it can also exaggerate errors while telling you that registration improved. From personal experience, there can be some topographic distortion from photogrammetric datasets, especially if GCP are not well dispersed within your area of interest. This distortion will make automated registration adjust your clouds to have lower (improved) RMSE while giving them a poorer registration for change detection.

A good test for any registration is to look for apparent spatial trends in your data using a cloud to cloud comparisons. For example, in cloudcompare, run a cloud to cloud comparison and use the slider bar in the properties to look for patterns of vertical offsets. Recently I had one that showed the center of my area of interest had the lowest vertical offset and the amount of vertical offset increased as I moved away from the center. My options are to rebuild the cloud, choose a spot within the cloud that I want to focus change detection on or raise my accepted minimum detection threshold.

If you use an automated registration to improve registration error (I do this frequently), it is a good idea to do it on a subset of your data that you expect to have not changed during the detection interval. That will give you more confidence that you are not "over fitting" your point cloud, or trying to reduce RMSE at the cost of reducing topographic change where it has actually taken place. In cloudcompare you can test the original registration this way by unchecking the rotation boxes and set the drop down box to "Z". In general the cloud will not move, but will give you the same RMSE estimation. You can use that value to see if allowing rotation and translation improves RMSE significantly (as defined by you).

It depends at what height you are flying at but in general for aerial applications (>100 m) the focus should be on infinite and to answer your question you should not have auto-focus. What would the reason of having auto-focus be?

In regard to the other setting, I would rather keep the shutter speed (1/1000 - 1/2000) and aperture fixed and allow for the ISO to vary based on the light conditions, if you have a decent camera it should not cause too much noise even in low light conditions.

Good luck!

thank you Reuben Reyes

what is your think of about software such as visual SFM or ....

thank you Marco Baldo

Hi When you do not have any of the RPC coefficients, you need to consider your work in several ways. One solution is to change the software used. I think you can use pci geomatica software and do it with DEM and polynomial rational function. The next solution is to use topographic maps either DTM or DEM to extract photographic control points. In this case you have to rectify your image with software like PCI or ERDAS.

Smaller UAV images in general (fixed-wing and multi copters) are quite different from classic aerial cameras in big planes.

Orientation angles are larger and their changes are more extreme .

Furthermore, UAV imagery often shows more motion blur .

and of course the ground resolution is much higher .

All this needs different strategies and matching techniques to be applied to successfully match tie points.

@ Stan: I have to admit that I am only interested in mathematical solutions which are possible :-), i.e. the ABC and the angles from D to A,B and C are meaningful values. Unfortunately, until now I do not find any analytical solution. Better said: I found one (from photogrammetry: Killian, K. Über das Rückwärtseinschneiden im Raum Österr. Z. Vermess., 1955, 43, 97-104, 171-179 ) which describes with 4 equations 3 variables (which I would call overdetermined system) but if derived (two angles) and inserted back in the 4 initial equations in order to determine the z-coordinate of D it delivers 2 slightly different solutions which can be optimized to a single one but this needs a a numerical refinement. I hope that I simply need an additional conditions or a totally different approach.

Yes, polarization filters vary constantly with the angle to the sun; the maximum affect is achieved at 90 degrees to the sun (photographers have a trick using their thumb and index finger). Linear polarizers are more effective than circular polarizers but most autofocus cameras that use through-the-lens focusing require circular polarizers. That said, for aerial SfM work at infinity focus distance, manual focus is preferred if you can fix (tape) the lens in place so linear polarizers can be used (they are also less expensive to purchase). Lastly, even if you use a circular polarizer, be sure to tape the adjustment ring so it doesn't rotate during your photo survey.

Abdullatif Baba and Aydin Ayanzadeh Thank you for the links.

I thought I would add a few papers which I have found:

Hi Matthew,

I don't know anything about this, but we had a guest lecture with this guy, Peter Jensen, who uses photos to make a 3D model of archaeological sites for documentation. I hope it helps a bit, if not, just ignore it.

Sebastian, recomiendo que uses MicroStation y TerraSolid y sus modulos. Puedes crear reglas de decision para limpiar la NDP...

Hi Dinesh, I use the Point Data Abstraction Library (http://pdal.io) for tasks like ground extraction. Check filters.pmf (https://pdal.io/stages/filters.pmf.html) and filters.smrf (https://pdal.io/stages/filters.smrf.html), there are links to the research papers describing the methods also. They’re pretty robust and can be used for mobile LIDAR applications.

I see you like MATLAB (based on tags) - you can expose MATLAB functions to PDAL as well (https://pdal.io/stages/filters.matlab.html#filters-matlab), although I prefer the Python API (PDAL can be called as a Python library as well as calling Python functions inside PDAL pipelines).

The neat thing about either approach is that you can use any object-finding methods developed in MATLAB or Python directly in the processing pipeline.

…and you can take advantage of point cloud library tools as well (https://pdal.io/stages/filters.pclblock.html#filters-pclblock).

I like PDAL because it’s easy to create recyclable processing pipelines pretty easily, especially messing around in automating processes, treating the JSON pipelines as templates which can have parameters passed in; or mixing JSON configurations and command line over-rides. And I don’t have to write my own file readers and writers (yet) :D

Extracting street signs seems like a cool project - keep posting updates; and good luck!

Thank you Martin, Florent and especially Reuben for these very helpful suggestions.

Advantages coming from photogrammetry are more or less understandable among professionals. But even among them there are those who treat orthophoto just like a picture, a nice background, not using it for any measurements. On the other hand, for an ordinary citizen orthophoto is the simplest way to do and understand things in the context of space, because almost everybody recognizes their home, road or anything else around them on the orthophoto.

Photogrammetric measurements are X, Y, Z, coordinates of your targeted points on the structure. First you must compute by photogrammetry Xo, Yo, Zo, the coordinates of the structure targets with zero strain, then aplying strain i compute by photogrammetry the coordinates of the same targets as Xi, Yi, Zi. Then the displacement will be Dx=Xi-Xo, Dy=Yi-Yo, Dz=Zi-Zo.

Hi. Generally this is a problem with Photoscan which happens when either you create a mesh with large number of points or when you create an orthophoto. In the first case try using chunks, NOT using Ultrahigh for your dense cloud or Using lower quality mesh. Alternatively you can create your points and bring them to Global mapper and do the mesh there. If you need texture maps you can upload them back to Photoscan and do the texturing then.

For the second, use tiles to create a set of orthophoto pieces. Then, bring them all into Global mapper and put them together in a format you need. Global mapper has no problem in doing so.

Dear Mohamed

Photogrammetry is already used in progress monitoring on construction sites due to its cost efficiency ( c.f. Scaioni et al., 2014). In addition, a practical example of its application of the integrated solution is merging 3D terrestrial laser scan data with photogrammetry for post asset inspection purposes (Monserrat and Crosetto, 2008). El-Omari and Moselhi (2009) have for example demonstrated improved site progress capture and monitoring with the combination of photogrammetry, laser scanned point cloud data and post processing software PHIDIAS. As these technologies continue to coalesce, the management of built environment assets will become increasingly reliant upon automated asset capture to fully exploit a vast array of structured geometric and semantic data extricable from a automatically generated 3D model (or mesh). Photogrammetry as such, offers to expand the current capabilities of facilities managers during O&M of assets within the built environment and in turn, improve business profitability and performance. Hopefully this has helped your query.

The following paper may be of interest: cy_of_optoelectronic_technology_developments_in_the_AECO_sector

Kind regards

Erika

Hi Matthew,

just a hint: if you want to deepen your understanding of this I recommend to read the book "Multiple View Geometry in Computer Vision" by Hartley and Zisserman, or at least some chapters about the basics of projective geometry. Actually, my explanations was inspired by that book. It describes everything very well, and is suitable for beginners, too.

Best regards

We currently have a project funded by Royal College of Chiropractors. Although we are developing a protocol using opto electronic methods, we can also use inertial motion sensors. This should help you.

looks like Pelobates cultripes (western spadefoot or Iberian spadefoot)

Best regards

Alfonso

Dear Tony,

I am an archaeologist and have published quite a lot on these topics. You can find all my papers here at ResearchGate. I have even grouped them in a special project: https://www.researchgate.net/project/Image-Based-3D-modelling

Cheers,

Geert

The resulting data from the photogrammetry is a 3D point cloud. This point cloud can be used to measure or extract 3D structures. To increase precision and accuracy added control points will help. The software for viewing and processing point clouds used were Agisoft PhotoScan, MeshLab, ParaView, CloudCompare, and LAStools. The point clouds can also be converted into DEMs that can then be used for analysis in ArcGIS or any GIS software.  Hope this helps.

Hi Javier,

If you are doing a parallel geometry (for example, building façades, inscription on walls, drone aerial photography), you can follow the classical formulas developed for large-scale aerial photogrammetry (for starters, look for a book called the Manual of Photogrammetry by Kraus and Waldhausl). You have formulas to calculate GSDs, number of photos to be taken for a specific overlap percentage, number of image bands, etc. I did that for a project concerning a building façade. This way, I can estimate the (minimal) number of photos to be taken and how much time I need to complete the photo acquisition. 

In order to adapt this classical method for a convergent geometry (e.g. statues, artefacts, persons...) I agree with Nicolas that it depends on your camera-to-object distance and the type of your sensor, although I would say 20% overlap is a bit too small... I always go for at least 60% in order to have multi-overlaps between the photos. Lower overlap percentage will also present a problem for the dense matching part, if you ever need to generate a dense point cloud.

That being said, the beauty of digital photography is that you can take virtually as many photos as you want. I always think that taking more photos is better than limiting them, as you can always sort the photos afterwards. On the other hand, on some cases you may not be able to return to the site to take additional photos.

Hope this helps.

Best,

Arnadi

Hi,

Thank you all for your answers.

all the best,

Mohammad

Dear Ebrahim,

It is important to tell what geomaterial do you intend for imaging.

For noise reductions, you can acquire images several times in the same condition of camera, and average their pixel values.

I suggest that take several images from each situation and intensity, and compare their corresponding results, and then set your system with that situation and intensity in which provide best results and go ahead for your next experiments.

Regards,

Am leaning towards pipeline which would use camera tracking program Syntheyes which would export marker and camera data to Photoscan for more exact tracking. Does anyone know of any research that has gone about doing this?

SEE THIS WEBSITE FOR YOUR ANSWER

PDF, Thanks to all. Regards

You can look at the following papers:

 Dear Nicolas

Thank you very much for your contribution. It has given a clear fundamenta idea regarding the subject. For an advance knowledge i will go through internet as per your suggestion.

Regards

thank you Sir, but i read this one before and i need more clear literature , esp. for the field of electrical high voltage equipment.

Let me quote late Professor Aleksander Pelczynski (one of the best functional analysts ever) from summer of 1972: "Important transformations from a Banach space into itself are only those which are close to the identity." He did not mention the nature of "closeness."

I believe it is a great opportunity for the dissemination of information and for new ways of making research. It is also fundamental in the sense of a "digital archive" of heritage. The risks are, I guess, not larger than the ones currently faced by heritage sites worldwide...

I may be wornd, but I don't believe that many people would stop wanting to go visit "the real deal" just because they can move/play/discover around in the world's most detailed reconstruction. They are different experiences and need not be mutually exclusive.

Absolutely not!

There is a profound difference:

The 2D projective transformation used in photogrammetry is not an one-to-one mapping or "bijecttive" mapping to use the precise mathematical term. As a mapping from points in 3D (object space) to points in 2D (photo plane) it maps all points on line passing through the focal point to the same image.

On the contrary the 3D similarity transformation is an one-to-one mapping from 3D to 3D.

Furthermore there is an equivalent to 3D similarity transformation for the plane, the planar 2D similarity transformation consisting of a single rotation, a translation with two components and a scale factor.

Your question: in general, compare the surveying engineering technologies for deformation measurement with the other tools such as dial gauges & LVDT.

I would say that:

1.     Surveying technology is 1D or 2D or 3D, as you wish; non surveying technologies usually provide 1D results

2.     Surveying technology can be related to a reference system located outside the influence of the object under study thus providing dimensions in an "absolute" reference system; non surveying usually provides relative measurements (for instance opening of a crack)

3.     Accuracy of these relative measurements usually are better (<1mm) than those provided by surveying technologies (usually accuracy above 1mm, exception: high accuracy spirit levelling)

4.     Surveying technology usually requires a specialized team (surveyors) to perform a surveying measurement campaign; non surveying instruments don’t need a operator specialized on that type of technology

5.     Surveying technology is a indirect method: you read angles, or distances, or photocoordinates, etc. etc. in order to provide the actual displacement (you have a function in between the observables and what you actually need); non surveying equipment provides, usually, what you need (for instance you read a relative displacement directly from an extensometer)

6.     from the points above you can conclude that you have two types of instrumentation for monitoring; these two types of instrumentation complete each other and check each other

7.     surveying campaigns are performed less frequently than those requiring directly measured observables mainly because the surveying campaigns are more complex and specialized in every aspect (personnel, equipment, processing, etc...).

I hope this helps...

Strictly speaking there is no "linear version of the direct linear transformation", this would be in fact a contradiction in terms. I presume that by nonlinear version you mean the original standard projective equations of photogrammetry. Despite its name the DLT transformation from unknown object coordinates and exterior and interior orientation elements is a nonlinear mapping (in fact a rational one). Linearity here simply refers to the fact that when the object coordinates are known, the DLT method model can be converted to equations which are linear in the exterior-orientation elements which are no other than the coefficients of the rational transformation (usually denoted by L1, L2, …, L11).

Now about the difference:

In the simple case of projective equations (i.e. the one without additional calibration parameters) we have 9 unknowns (3 coordinates for camera position, 3 angles for camera orientation, 2 coordinates of principal point plus 1 focal distance).

For the DLT method to be exactly equivalent to the projective equations one must incorporate (11-9=) 2 conditions on the coefficients. These conditions however are nonlinear functions of the DLT coefficients and this destroys the advantage of using linear equations and thus avoiding the approximations of linearization in order to use linear model (least squares) methods fo data analysis.

For this reason the 2 conditions are not used in practice. The 2 additional parameters in the DLT method are accounted as additional calibration parameters. However there is no clear projective equations model with 2 additional calibration parameters, with its 11 unknowns being explicitly related to the 11 DLT parameters by well define mathematical relations.

In conclusion the DLT method can be considered as equivalent to the projective equations with implicit additional 2 parameter calibration.

If you send me your email I could send you my class notes (in Ebglish) from a course I tought last year at the Milan Polytechnic, where the above are explained in detail.

Best regards

Kite are already used for coastal monitoring. You are right, it is an inexpensive method. However, I agree with the previous comment: they are difficult to control and sometimes crash. And if developping, this activity might be regulated.

Coastal monitoring is not only beach monitoring... What about cliffs ?  You can pilot a drone from the top of the cliff, the drone flying at lower altitude than you. This is not possible with a kite.

Moreover, some beaches are surrounding by cliffs and therefore the flight plan of the kite may be restricted...