Classic applications for object-detection are many and varied, ranging from looking for items such as wreckage, debris, lost items and so forth to defence applications, seeking out threats such as mines. Increasingly, though, object detection has been expanded to include obstacle avoidance for both surface and submersible vehicles, and an expanded defence and security requirement to search out additional threats from mines and explosives in and around ports, harbours and marine installations. Despite technological advances in imaging and mapping, when it comes to object detection in water the humble sonar remains a vital tool and is used almost exclusively for imaging and mapping of the seabed or structures such as harbour walls, platforms and pipelines.



When we refer to 'object detection' in this instance, what we really mean is finding specific targets, usually lost against the seabed or amongst structures. In reality, all sonar detect objects; indeed, the only thing sonar does is tell you that something (i.e. an object) is present. In general, as far as imaging and mapping sonar is concerned a lack of anything in the water will result in no data, or at best a uniform, featureless image.



Existing Techniques
Notwithstanding laser scanning, which can achieve extremely high-quality results but only at short range, sonar is the only practical option for underwater visualisation. The problem with sonar is that it is not easy to understand or interpret and therefore object detection requires a high degree of skill, experience and plenty of luck. For many years, side-scan sonar and sector-scan sonar have been the mainstay of imaging and object detection.



Classic images seen in publications and marketing literature show stunning images of shipwrecks and the like; however, less obvious objects return more cryptic sonar results. Side-scan and sector-scan sonar has been likened to shining a torch in the dark; it is possible 'see' what is obvious and directly ahead, but obtaining good coverage, sufficient to make a detailed analysis and interpretation, is a tough challenge. Often a significant proportion of the information available to the interpreter lies in what cannot be seen, i.e. the shadows. Skilled sonar operators and geophysicists are able to make accurate and useful analyses, but even the most experienced interpreter working with the best data can make mistakes.



The following slightly anecdotal example illustrates just how difficult it can be to interpret sonar data, and equally how easy it is to make a mistake. One particular case of the many wrongly interpreted sonar images surrounds the location of a wrecked steam liner in the Indian Ocean. Using the latest side-scan sonar and employing the most fastidious survey techniques, a highly regarded surveyor produced incomparable images of the wreck to compare against photographs and drawings of the liner. It was possible to match practically every feature on the sonar data to parts of the ship and to draw the reasonable conclusion that the lost liner had been found. In most cases the story would have ended there, unchallenged or corroborated, everyone involved none the wiser. However, in this instance, manned submersibles and TV film crews were deployed to further investigate the wreck. On reaching the site it was immediately clear that it was not the wreck it was first thought to be, but in fact a similar vessel used as a Second World War tank carrier. On reviewing the data to determine how this mistake had been made it was noticed that one feature of the steam liner, a stairway on the stern, did not exist on the WWII tank carrier, although it showed clearly on the sonar images. Further investigation showed that the tracks of an overturned tank appeared at almost the exact spot where the phantom staircase had apparently been detected.



Since the advent of digital signal processing there have been many attempts to automate the object-detection process but few if any have proved reliable; they give variable results and are not widely used. The primary problem is the variability of sonar data as a result of height above the target, direction of survey and so forth. Consequently, like many similar tasks, object identification is best performed with the human eye. New and better imaging techniques need therefore to be provided from which to make more accurate analysis, at least until automated object detection and classification becomes a reality.

3D Sonar Alternative
Originally developed for the sub-sea construction industry to aid tasks such as well-head intervention and welding where semi-automated control systems need precise information about the immediate environment, updated in real time, 3D-sonar is a relatively new sonar implementation. Developed over the past ten years but only recently become commercially available, 3D-sonar, as the name implies, generates 3D images. Unlike conventional sonar such as multi-beam and scanning sonar, which can generate 3D images by scanning and combining data, usually as a post-process, true 3D-sonar creates 3D data instantaneously and in real time. A 3D-sonar image is created by simultaneously forming acoustic beams to fill a 3D volume, as opposed to a narrow fan of beams. In the case of the Coda Echoscope, one of the only commercially available 3D-sonar, over 16,000 beams are formed, each yielding a data point or voxel (3D pixel) with a position (X,Y,Z) and intensity (i).



In simple terms, a 3D-sonar can be thought of as 128 conventional multi-beam sonar bolted and squeezed together to produce 128 x 128 beams at the same time. By plotting the data points on a screen in much the same way as 3D games, a 3D rendition of the underwater scene is produced. Because the data is three-dimensional it can also be moved and rotated to change the perspective for better visualisation and interpretation. The advantages of 3D-sonar over conventional 2D-sonar are several:

• each individual 'ping' yields a complete image without the need for scanning, motion compensation or post-processing


• by scanning and combining multiple pings, much higher coverage is achieved, including a high degree of additional in-fill from non-orthogonal angles


• stationary 3D-sonar can be used to visualise a moving scene.