Last year I experimented with some photogrammetry in the confines of the Daren Cilau entrance crawl, the section known as the vice to be more specific, details can be found here:
Spurred on by the success in this short bit of passage I convinced myself that it would be a worthwhile project to try and accurately model the entire crawl from the surface to the final rescue box. One could simply video the entire crawl in single trip and use the video to produce a model but it is important to me to make this model accurate and to do this it needs to be surveyed so that the resulting model and can aligned to the survey data. For assessment of accuracy I surveyed both in and out so the loop closure could be examined and any poor data identified and sorted. To make this task more manageable I have broken the cave up into sections between the rescue boxes, these also make handy survey markers that aren’t likely to change location in between trips.
To date I have completed three trips; a trip consists of caving to the rescue box at the far end of the section with a small bag, surveying out from the box to the previous one then surveying back in. Caving slowly out with the Gopro and light trying to get steady footage before turning at the previous box and filming back in to the starting point. I then pack up the gear and cave back out. This involves lots of awkward caving for those familiar with the cave.
Home made ‘filming stick’
The first two of the trips have been successful and a decent model has been created from the surface (box 0) to box 2. On the third trip I found that my Gopro had been switched on prematurely and battery warning was already on before I started but this wasn’t discovered until after complementing the survey so I had a rather rushed trip filming back from box 3 to box 2 before the battery ran out. After some failed attempts at saving the data from the third trip this means I will have to repeat it as I was moving too fast for the image alignment to be successful.
Plan view of disto data from entrance (B0) to box 3 (B3) coloured by section/ date surveyed.
Its has taken some time (a bit longer then planned) but i’m pleased to say that the entirety of Ravens Well has been surveyed and photographed resulting in a complete and accurate (to the disto data) model. The same techniques and equipment that were previously mentioned have been used, some sections proved tricky for the software to understand so I have introduced the coded Metashape markers to provide visual and spatial reference, this seems to have solved the issues in certain sections of passages where the alignment was struggling.
Coded Metshape Target
There are some ferrous areas inside as the disto data shows some larger than expected loop closures in certain areas despite careful practice. Metal girders are visible in the far downstream area which is the bit which shows the greatest variation when overlaying my data with the previous survey performed by Axbridge Caving Group.
Plan view of the Distox2 data in Survex
I’m working on a video walk through of the whole site like the previous version so I will upload this when its complete, here is a quick video overview for now, below is low resolution model uploaded to Sketchfab which you can explore with your mouse/ phone.
Video Overview of Model
Low resolution Sketchfab upload (Due to free account limits)
A few friends have been working at Rickford Rising removing large boulders and making progress downwards from where the previous protagonist left off some years ago. I have always fancied a dive here but had never got round to it for various reasons. It usually has good visibility so I thought I would attempt to create an accurate photogrammetry model of the site as a snap shot of progress at this point in time, and hope to go back and update the model as progress towards the Burrington Master Cave is made !
A video has been put together by the digging team and gives a rough overview of the site:
I made four short dives on my visit, facing head first slowly filming the descent from surface to dig face allowing sufficient time for the visibility to clear each time and a survey dive (after having planted the yellow builders square on the first dive).
The builders square was used to provide a scale reference (the lipped edge is 12″ or 0.31m long) and also to provide a fixed straight edge along which I could repeatedly align the edge of my survey box with to make foresights and backsights against to allow the model to be orientated correctly. Accurate depth of the square was also measured, 6.8 m to the yellow surface on the day but this will vary with flow and water level effects.
End of the dig and planted builders square (krab is ali)
I have setup the survey box so that it logs data continuously, this allows the diver to retreat whilst measurements are taken minimising any magnetic effects of steel cylinders or other dive gear, this is apparent in the good agreement between foresights and backsights taken, they are no worse than what I have measured on land away from ferrous materials. The upper part of the dig is festooned with scaffold and other metal things.
Survey results
The survey data was input to Survex and corrected for local magnetic deviation, a resulting bearing of 161° to was calculated for the lipped edge of the plastic square.
The model was processed without issue and was manually scaled, rotated aligned and translated with the survey data measured and the entrance location provided by the cave registry in CloudCompare.
The video from one of the dives can be seen here, attempting to slowly film the waters surface from underneath, down the rift to the current dig face.
A flythrough of the model is available to view below:
The image above shows progress so far, colour coded by area
Over the summer months this year (2021) I continued to visit, survey and video various parts of the cave upstream of Upper Cave Water Chamber. The results are very pleasing because not only is the level of detail being captured far in excess of any survey that has been conducted before (at this site) but this detailed data (the models produced by photogrammetry) is also matching well with the Distox2 data which gives real world scale and orientation meaning its not just pretty pictures which I prefer to avoid.
This area of the cave lends itself very well to my process as each section is short and clearly divided by short sumps giving nice workable areas to focus on. The below image shows a comparison with the UBSS survey in plan view.
Historical comparison (UBSS on left)Plan view, blue line where visible is the DistoX2 centreline data Side view, blue line where visible is the DistoX2 centreline data UCWC looking downstream from Sump 8 (textured model) Cobley crawl looking downstream into Sump 8 (textured model) High level muddy tube (textured model) Sump 9 looking upstream (textured model)
Where to go next ? I would like to improve the alignment of Sump 9 relative to the rest of the model as so far this has just been best fitted to the other data and I would prefer to link it directly. After this then adding the passages in Parker Series is the next logical step but perhaps this will have to wait until next summer.
I have uploaded the models of the three dry sections to Sketchfab, its low resolution due to the limits on the free account.
So far my attempts at underground photogrammetry have been limited to easy places where walking steadily through a passage poses little to no issue, this where caves are involved is somewhat limiting due to the nature of cave passage formation.
I set myself the challenge of attempting to model something smaller and less easily walkable than what I have done so far to see how collecting the video might be and how well the software would cope with what will likely be non ideal footage for reconstruction.
The entrance passage to Daren Cilau struck me as a good place to try, more particularly the section known as the Vice; A narrow section in which cavers must turn onto one side and keep their bodies high to prevent being jammed into the narrow trench in the floor. It doesn’t pose a particular difficulty to myself in terms of the passage vs my body size and fairly large people can get themselves through but I thought it would make an interesting experiment, the main issue being capturing reasonable video with one hand whilst negotiating the section of cave. It is an iconic cave in British Caving and probably has never been measured before in such detail.
With the above goals in mind I set off one day over to Wales, my plan was to cave the short distance into to The Vice, mark some (min of 4) stations on some prominent features before, through and after The Vice, survey these stations with a distoX2, then cave through this small section a few times whilst trying to take steady video. Using the marked stations the resultant model could then be scaled and aligned correctly. These stations were marked with Tippex which was carefully removed after use.
Source video of one of the outwards trips
A single dive torch was used attached to my Gopro on a cheap ebay selfie stick to keep weight and bulk down.
In practice this went fairly smoothly, the Vice was passed a number of times: In: Marking stations Out: Survey with Distox2 In and out: filming In and out: filming again In and out: removing Tippex stations
On arriving home I set about processing the images and compiling the survey data. It has become my normal practice to film both directions into and out of a passage, if the alignment is to be considered accurate then both in and out passes should match up (plus any additional passes). This is closing the loop which allows relative accuracy to be assessed and also gives a different perspective on features if its successful.
Separate models for each in and out pass were created first, unfortunately each one giving a slightly different representation of the passage when attempts at aligning them in Cloud Compare were made. I then made a model combining both a single in and out set of images, this was successful in that there only appeared to be a single cave passage which showed that the software was able to combine both sets of images semi correctly. However when this point cloud was matched with the survey data although a low RMS (0.05 m)value was reported the survey data appeared to shoot through the walls of the cave which I know not to be true. I then made a new attempt using all in and all out passes and matched the reference points with the survey data again, this time I got a low RMS (0.03 m) value and the survey data appeared as it should within the bounds of the walls, a good result which I am happy with. This goes to show the importance of some quality control and verification against other sources of data. The images below detail these results.
Single inward pass fitted to survey data
The above image shows the model created from a single inward set of images, Metashape reported all images were aligned and after 7 parameter transformation to the survey points in Cloud Compare a low RMS value of around 0.05 m was reported, however on inspection the survey data (blue line) goes through the walls so something isn’t right. The deviation occurs through the narrowest section where steady filming was most difficult. The image below is the result of the 7 param. transformation.
7 parameter shift results for the single inward model
The below images show the same process as above but using the model derived from a combination of all 4 in and out passes.
All in and out passes fitted to survey data (the roof cut off to allow the data to be seen)
The results for the 7 parameter shift can be seen below:
7 parameter shift results
I am very happy with the results of this, not only has a model of the cave been constructed but it also appears accurate when aligned to distox2 data. The model plus a few extras to aid with scale have been uploaded to Sketchfab here:
In the mean time I have added some survey data with a Distox2 to align the model to giving it scale in meters and alignment to magnetic North. The results were quite pleasing from a surveyors point of view aligning to with 0.2m for the 4 points used.
Alignment was achieved using a 7 parameter transformation in CloudCompare meaning the model is now more than just a visual representation. Work is ongoing to complete a model for all passages at this site.
A trackplot for the flythrough was created and added as an overlay to the video to give an indication of location at any given time, this was done in DashWare, an excellent free video editing program.
Porth yr Ogof is a cave situated in the Brecon Beacons in South Wales frequented by divers as well as dry cavers, I believe it has the largest cave entrance in Wales.
The large bedding plane entrance to the main cave
The naming convention is rather confusing in that to reach sump 9 you start in sump 3 at the Tradesman’s entrance, dive through sumps: 3,2 and 1, before exiting the water for a short section of passage (Upper Cave Water Chamber), before diving sump 8, crawling for a short while before diving sump 9, the subject of the post. It is also possible to start in sump 1 entering via the Top Entrance negating about 150 m worth of diving by missing sumps 2 and 3, information and an old survey can be found here:
Late in 2020 I visited Parker Series which is the dry section of the cave guarded by sump 9, after diving through it struck me how ideal a place it was to attempt to create a model using photogrammetry. The water is usually clear and the passage is mostly formed as a small tube meaning cheap lights and cameras would make a reasonable job of it.
Source Images from first attempt
I set off on a second trip with my Gopro Hero 3+ and Ebay video lights and moving very slowly captured stills every 0.5 seconds using the interval function. It took around 300 images to cover the sump each way. The Ebay video lights I used do have an odd colouration where they overlap but this doesn’t bother me, it might bother photography purists but my main intention is to measures the size, shape and direction of the passage and for this true colour representation doesn’t matter.
Typical source image, taken about midway through the sump
Loading the 300 images into the software and processing the images yielded a nice accurately aligned point cloud of the sump, I didn’t cover the three side passages on this occasion so these are missing but the smooth curves that the passage follows can be clearly seen, the UBSS survey lacks detail in comparison.
Plan view of model
Whilst a nice representation of the sump has been created it bears no orientation or dimensional reference to the actual passage. I visited again with the intention of covering the side passages and adding a means of aligning the data to magnetic north, absolute depth and scaling it so that distance and size was more accurately represented.
To do this I took a plastic builders square (with weight attached) and my survey device back to the sump, I dived a short way into the sump before placing it on the floor and returning to the start for my camera. I then took photos through the passage as normal this time covering the builders square on the floor. I continued to the first side passage and tried to cover as much as possible of it. Once this had been done I placed my survey box on the raised edge of the square to be used as a heading reference and took 2 for-sights and 2 back-sights. I had programmed a timer so that I had time to retreat a few metres for each shot so as to minimise any magnetic distortion from my diving equipment being close to the magnetometer. The fore and back sights agree quite well so it was worth doing this.
Plastic builders square with survey box used to measure heading and depth.
I repeated the process towards the end of the sump so that I had two heading references (one to align to and one to check against), the depths of the centers of the yellow square were measured to align the depths to and the edge of the yellow square was used to estimate scale from. Once out of the cave the photos were then processed to generate a new point cloud containing the yellow squares which were then rotated, translated and scaled to the ‘correct’ figures.
Well here we are again, in the throws of another lock down. Travel from your local area is frowned upon, all venues (apart from the sea) are closed for diving so to keep stimulated both physically and mentally I decided to experiment with some underground (local) photogrammetry.
My experience of photogrammetry is limited to mostly following others around underwater whilst they photograph things, a few work projects and some failed underwater attempts in a cave (perhaps a separate post on this later).
For anyone thinking of taking it up I have had very good results with mobile phone cameras, Gopro’s and cheap lights, the process is very simple to perform on a computer so I would encourage people to have a go, the software can be trialed for free so expensive equipment is not necessary.
So back to the lockdown; fortunately I have a site I can access within walking distance of my house in which I can experiment, it is an old water conduit known as ‘Raven’s Well’
Armed with a set of cheap waders from Ebay (its waist deep in some places), a Gopro hero 3+black and two cheap video lights I set off to capture some photos to see how well I could model a part of the site.
I set the Gopro to take a still image every 0.5 seconds, put the lights on full and set off walking slowly around the passages near the entrance with the camera pointing forwards. Care at turns was taken to ensure that lots of overlap was achieved. There is a loop that can be traversed so I walked around to see if the software was able to accurately ‘close the loop’ a fundamental part of survey data assessment.
I went round the loop twice in an anti clock wise direction before heading downstream to the low section before returning to the loop and completing the loop again twice in a clockwise direction, this amounted to 1237 photos, just over ten minutes of photo capture. I have collated them into a short video so the quality and coverage can be seen.
Source Images
This amounted to just over 4 GB of data, the details can be seen be below for the Jpeg images for the photographically minded.
Source Image details
Images can be harvested from video but they lack the metadata that comes from still images so I find this approach easier provided you take enough images first time around, with video you can extract more frames without revisiting the site if required.
Photogrammetry is a computer intensive exercise so before I pressed the ‘Go’ button on the whole set of images I tested a single loops worth to see if what I had captured was going to be worth the wait for processing, this took about two hours to go from raw images to dense cloud, I was happy with the result, it failed to close the loop but had modeled the shape and course of the passages very well, see the below image.
Results from single traverse of loop
The above image is a plan view of the dense point cloud created from one walk round the loop. The areas circled in red are the same physical areas and should join up however at the area highlighted with a blue line (the first corner) it has failed to adjust for the camera heading change properly which can be seen by the ghost walls, if this piece is manually cut and swung round it allows the areas in red to overlap. I was encouraged enough by this to select all the images and pressed the ‘go button’. After all I had 3 more traverses of the loop and hopefully the addition of more images would help it close properly.
This was a much longer process, which took around two days (Macbook Pro running Windows 7, 64bit, 16 Gb Ram, i7 2.9 Ghz). Waking up to a silent laptop (the fans goes into over drive when its processing) on the second morning I was pleased to see all images had aligned and it had finished so I loaded the dense cloud and started to inspect it. I was very happy with the results, the loop had closed and the passages appeared as they should. The image alignment was run on ‘Medium’ and the Dense Cloud was set to ‘Low’. More detail could be processed at the expense of processing time but for me this is good enough.
The result of processing all images
The first job once the initial overview had been completed was ‘cleaning’ the water out of the floor, most of the areas have a wet floor and its unsurprising that it struggles to model a constantly moving, colour changing body of water so these points were manually selected and removed. Once this had been completed the mesh and texture were computed, taking just a few hours. Below are some selected views from inside the model, I am working on some sort of video or fly through to be posted when available.
Looking towards entrance down brick roofed tunnelLooking downstreamLooking up the inclined section
Future work will involve covering the rest of the site and geo-referencing the data to the real world as an arbitrary scale and alignment is applied straight from the software.
An internet search will bring up a number of devices people have created to help when calibrating a DistoX2, mostly these seem to require the use of a 3d printer or a trip to the local plumbing store. I have made a rotary cradle out of some scrap plastic material that allows the the heading of the Distox2 to be maintained whilst rolling the unit to collect the required shots for calibration to be performed.
DistoX2 mounted in plastic jig
It is two discs (cut on a lathe whilst sandwiched and bolted together so they are exactly the same size) which clamp around the outside of the Disto body. The front has a large hole for the laser and the rear has a small hole in which the rear reference point is to be aligned with.
Front end with clearance for laser beam, rear hole to align with reference point
Before use the front end (whilst maintaining the position of the reference point to the rear) must be moved so that the laser exit is inline with the centre of the circle of plastic. This was done via trial and error, the jig and Disto was pointed at a wall around 4 m away and rotated until the laser described as small a circle as possible (nearly a dot) on the wall.
Once satisfied I then collected the 56 shots required for calibration. I prefer doing this in my garden, i’m yet to get better results using targets on a cave wall.
Using a plastic mitre to keep heading constant for flat shotsNon metallic objects used for angled shots wood held together by dowels not metal !
Once the shots have been collected they are grouped then analysed in Topodroid, i’m very happy with the results and would definitely use this jig the next time I calibrate the disto
The launch of the 5th edition of Mendip Underground caused somewhat of a surge of interest in certain caves in which it contained updated descriptions, photos and rigging diagrams, Mangle Hole near Sandford being one of them. It was the rigging topo which caught my attention as it suggests some nice free hanging pitches with multiple rebelays and a sump at the bottom, what more could you ask for !
The reality is somewhat different, its reputation does seem to precede it and it is seldom visited but I quite enjoy a trip there and have undertaken a few club trips, several digging trips, several surveying trips plus a failed dive trip.
As there was no decent published survey of the cave (and there still isn’t) I decided with my new found cave survey skills I would have a go myself. I was joined by Pete Hall and we set about surveying the cave, I don’t remember much detail of the trip itself but I can remember feeling quite frustrated when sitting down at my computer with the data afterwards knowing what shape the passages and chambers actually were and seeing the shapes that I was to draw on the survey to represent them.
The problems lies in that the cave is very steep in nature, the entrance rift is very narrow floor to ceiling but wide in the wall to wall dimension, the chambers are large and vertical in nature with lots of alcoves and side rifts, the route to Aldermaston Chamber is small, complex and muddy. When the splay shots are viewed in plan view the whole cave looks very different to the reality.
Attempt at Plan view in CAD, Splays in yellow, survey legs in red.
At the time I was trying to learn how to draft surveys in Therion but perhaps this multilevel cave was a bit to challenging for my basic skills. Instead I found a handy export function of PocketTopo where the raw data is exported as a 3d *.dxf file, this can then be loaded in CloudCompare where it approximates the walls based on splay shots and shows a 3d model which can be spun around and viewed and understood more readily then a plan and elevation view.
One of the exquisite mud formations, this one looks like a mans head
