Back in the Water Part II

The imposed layoff from underwater photogrammetry has finally lifted. Compared to what others have faced during the pandemic its low down the list of issues but after 6 months of no camera work it was time to dust off the kit and brush up the skills.

The first dive at Vobster was a navigation exercise using the UWIS system with the joint mission of validating the marks from previous dives and then locating and marking the drilling/piling rig which sits isolated and off the guide line.

The second dive plan was simple: return to the rig by shortest route followed by a wander out into the abyssal plain to plot some targets.

The yellow line represents the first dive route, and the orange track shows the (almost) straight route taken back to the marked rig.

Screenshot showing the Valtmer navigator software running on the Alltab with a series of marks plotting the location of dive attractions. Background is a DEM derived from multi beam sonar data and supplied by Max and Geodesea Ltd.

Following user feedback we changed the DEM colour palette, dropping the mark-hiding green.

Camera Issues

After a near straight line navigation back to the target a slight flaw in the plan appeared; a camera button had jammed down.

On Nikon cameras this does not prevent images being taken and the basic settings like ISO, aperture and shutter speed work but it does shut down the rear preview screen. The last time this happened was back in 2007 on the SMS Markgraf…one image ended up as a double page spread so nothing was lost…so it was time to apply a semi-educated guess for the camera and flash settings.

In spite of camera issues just under 300 images were shot and aligned. Image courtesy Rick Ayrton.

Processing Pain

The results turned out OK. The drill does not sit in the pool of light preferred style but all images aligned with scale and accuracy (key goals) were tight. Up to generating the mesh everything looked just fine including the long/thin vertical poles – features that can pose their own challenges.

When it came to processing the mesh issues appeared.

When it comes to generating a mesh we have two methods; the traditional dense cloud sourced mesh and the more recent depth maps method.

Depth maps offer significant advantages over using a dense cloud – we covered that in more detail here – and the dense cloud is just not processed. In this instance the depth maps derived mesh was…less than idea:

The less than ideal results with mesh growth appearing around the wheels

The issue here has been triggered by a lack of camera coverage.

The wheels have a recessed centre leaving an underhang and with any gap the mesh will start to grow and fill in the missing details. In this case the mesh confidence reveals the missing areas, with red indicating limited coverage and blue showing strong confidence:

Model confidence – available with depth maps – highlights the areas of pain (in red) and confidence (blue). Lack of coverage triggers this issue.

Now before we go too far and start kicking ourselves in the backside we need to remember the quarry bed can be a little silty at times and getting the camera into the tight spots might cause a bit of an unwanted fog. But the main trigger was twofold; a rusty photogrammetry mindset compounded by distraction triggered by camera issues.

Step Back

When working with closed and seamless camera coverage depth maps work very well. But when the source data is compromised we are not tied to a single method and generating a dense cloud for mesh source helped resolve the issues:

Practice & Prepare

Quarry diving is the perfect opportunity to refine the skills without the pressures of boats and tides. This applies to diving and photogrammetry skills alike and keeping both fresh is preparation for what is to come and “check the image preview” along with “check for flash trigger” has been embedded (again) into the pre-dive camera prep.

And when things go less than perfect remember Agisoft Metashape is a toolkit designed to deliver many things, but the same tools are there to resolve issues and provide workarounds for issues the developers never even dreamed of – it goes without saying training helps with this no end.

There are a lot of sonar targets waiting in the English Channel and being ready just loads the dice in our favour.