Rovco’s SubSLAM X1 camera creates real time 3D models
In the digital world, information is currency. But some bits of information are more valuable than others. Getting high quality data on subsea infrastructure is worth operators spending thousands on inspections.
Subsea services company Rovco has developed a 3D scanning tool to make the process easier. At this year’s Subsea Expo, the company revealed a working commercial version of its SubSLAM X1 camera. Mounted on an ROV or AUV, the camera can reconstruct an asset in 3D in real time.
SubSLAM X1 uses a 4K stereo camera unit to collect a series of static pictures at up to 30 fps. The unit’s algorithms spot key features to make a millimetre-precise 3D point map. Moving elements, like fish or sediment clouds, are not captured. SubSLAM X1 can work at depths of 1,000 metres, with deeper rated systems planned.
The data is stored internally and sent to surface live via copper, fibre or ethernet cable, before being transmitted back to shore on a 1MB link via a ship’s VSAT system or a platform’s 4G network. As the map is built up, layer by layer, the operator can see when enough information has been captured, or whether to keep the vehicle in place to gather more data. The high resolution model can then be analysed onshore or processed further.
Rovco CEO and founder Brian Allen told InnovOil at Subsea Expo that he knew of nobody else that is capable of currently producing accurate, live 3D underwater.
What appears to set SubSLAM X1 apart from other survey tools is both the speed at which it gathers data and its accuracy. The live map has an error of less than 0.67%, under a centimetre in every metre. Processing brings this figure down.
The 3D model can then be used for difference modelling over time, comparing one set of data with an earlier set, providing data that can help predict the lifespan of a part. “For instance, you can see how your anode is depleting through difference modelling two data sets, giving you the depletion rate.”
By collecting data in real time SubSLAM X1 removes the post-processing step. “It’s coming straight to the surface in a useable format, so our cost for data delivery is lower.” This speeds up projects, cutting the time needed for vessel rental and crew support.
Like all cameras, SubSLAM X1 runs into problems with poor visibility, less than one metre. “The ideal range is five metres,” Allen said. “We can work from 0.3 metres back to 10 metres and still get good data.”
While any visible object can be reconstructed, in murky or deep water visibility is unlikely to reach more than eight metres. This, however, is common to all camera-based systems.
Visual processors have trouble with reflective objects, which will appear as inaccurate shapes or be missed altogether. SubSLAM X1 uses polarising lenses to kill the shine.
SubSLAM X1 is resistant to currents and tides as they have minimal effect on its scans. The software builds the 3D point cloud piece by piece. As long as the operator stays close to the object being scanned, a picture will emerge in less time than a conventional static camera would take to take a single “perfect” picture. All the data goes to form a larger picture, so less time is wasted pointing the camera in an unproductive direction. If the camera is diverted by water movements models simply pick up from where they left off when the ROV re-orients itself.
The camera can be mounted on a pan and tilt mount on work-class ROVs, so it can hold itself in a current while the camera records at a right angle for extra stability.
Since SubSLAM X1 only tracks static images it has trouble when scanning moving parts of a scene that it is not trained to ignore, such as a loose cable flapping in a strong current. “For example, an electrical flying lead in a really strong current could be captured twice.”
As almost every object relevant to a survey will be protected against the environment, most moving parts will not need to be mapped. Should new technologies, such as tidal turbines, become a commercial reality, SubSLAM X1 will need to adapt to account for this.
Technical innovation often fails when it gets out of the lab and in the field, especially underwater. Rovco has anticipated some of the interpretative challenges SubSLAM X1 will face, like keeping a firm frame of reference when the ROV is lost or spun quickly.
Rovco has added a localisation system to the code, which makes SubSLAM X1 regain its bearings by picking out a single static piece from the scene to orient its software.
“We modelled our office Christmas tree as a test,” Allen said, “because it’s particularly hard, with lots of small details and shiny elements. Someone shook the Christmas tree and it changed enough for the camera not to recognise it. When SubSLAM loses localisation its model stops developing – it doesn’t recreate a second Christmas tree, but reports that there’s been a localisation issue. Then, after a pause, it restores the scene before the disruption and continues mapping.
If SubSLAM X1 loses complete track of its picture (perhaps because a fish that was stationary moves) the software steps back 10-15 frames until it regains localisation.
Despite being a fluid environment, the seabed is relatively static. Major changes are rare and slow moving, and are unlikely to lock up the system completely. At worst, they will delay data capture for about half a minute. And since the data is being processed live, any errors will be obvious.
Allen believes that SubSLAM X1 is the first building block of an AUV. Should the visual system recognise an object, the ROV would alter its behaviour accordingly. However, that goal is obviously some way off, but the team have UK government backing via Innovate UK to deliver that reality.