98 In operation | IQUA Robotics SoundTiles As the system functions independently of navigation data, it can generate mosaics of FLS data in any survey pattern; not only horizontal surveys, such as lengthy tunnel or river mapping or conventional back-and-forth zigzags over a broad area, but also vertical inspections of harbour walls or dams. This makes SoundTiles valuable to civil engineers and port authorities, as well as energy industries, environmental agencies, researchers, and many other groups. User interface The general pipeline of the user’s experience of SoundTiles is as follows: they collect their sonar files, open SoundTiles, create a project from their sonar data, and are then guided by a wizard through the steps of generating their mosaic. In most cases (particularly if the survey achieved reasonable quality data) the software outputs a sufficiently clear mosaic and the process ends there; in other cases, some additional tools are available in the software for fine-tuning the result until the user’s ideal mosaic has been created. When starting a new project, the user selects their sonar data file and a folder for the mosaic output, and they can also pick a navigation data file; for instance, if they feel it may be helpful to their project despite not being necessary. Next, the user is presented with the mosaic wizard’s initial screen, which summarises a plethora of information immediately extracted on the dataset (such as the sonar model used, the date of the survey, and the total number of images or frames in the file) and on the sonar (including the field of view (FoV), frames per second, resolution and pixel size). The user can also specify a few other options in this screen, such as whether the survey was a simple rotational one (with the UUV panning horizontally on the spot), a conventional ‘constant heading’ zigzag pattern, or another. They can also choose whether positioning data will be overlaid in the mosaic, used as a primary guide for the mosaic (although, as one may guess, this is not recommended unless one intends to compare the SoundTiles mosaic with a conventionally produced one for relative quality) or disregarded. A preview screen is also generated here, with a slider for the user to quickly flip through all the sonar images in their file individually. Users then drag sliders to choose which frames to process into the mosaic: when a UUV starts or finishes a survey session outside of the mission-critical inspection area, eliminating those extra frames saves time and computing resources. They may also specify individual frames to be discarded, as there may be multiple overlapping and therefore redundant frames. After this, the software can generate the mosaic. Registration and generation The first step in mosaic generation involves the software registering or aligning the sonar images together. The time to do this varies, depending on computer processing power, and the number and size of the images, although it typically runs in the order of minutes. Hurtós notes that with a small dataset on a high-end PC, it could take just 10 seconds. “There’s no need for any GPU or fancy processing unit, as the software is configured to run off the user’s CPU – and the more cores their CPU has, the more it can parallelise the image alignments, so the shorter the time it will take,” she adds. During this time, the algorithm tracks the sequence of images collected along the trajectory and timing that the UUV followed, and it establishes windows within which consecutive images can be registered with each other; the user is encouraged to set window sizes of three to five images. “Because we’re creating a 2D map, we can ‘register’ images with each other June/July 2024 | Uncrewed Systems Technology As well as stitching together mosaics of the sea floor, SoundTiles can produce mosaics of vertical structures, such as this vertical wall surveyed at a flood barrier using a Kongsberg Discovery M3 acquired dataset As the system functions independently of navigation data, it can generate mosaics of FLS data in any survey pattern
RkJQdWJsaXNoZXIy MjI2Mzk4