underwater researchのための3Dイメージングの進歩

タイトル: Influence of camera type, height and image enhancement on photogrammetry success in turbid marine environments.

概要: Over the last decade, Structure-from-Motion (SfM) photogrammetry has successfully been used to survey marine benthic environments, including artificial reefs, shipwrecks, and coral reefs, for a wide range of applications. The method is likely to become one of the most common tools for surveying marine benthic environments. However, SfM photogrammetry has been developed in clear water environments, and its suitability in turbid, benthic environments is less certain. Turbid coral reefs are example of an important marine benthic environment, making up 12% of coral reefs globally. Corals in these environments have a tolerance for low-light and high sediment conditions. Such attributes mean these reefs may be important refuges from extreme light and temperatures. Therefore, assessment and optimisation of the photogrammetry methodology in these environments is needed. This study investigates the performance of SfM photogrammetry in turbid environments, by comparing two camera types, settings (automatic vs manual derived from local conditions), the height of image acquisition above the seafloor, and post-processing image enhancement. Three dimensional (3D) SfM photogrammetry meshes of an artificial reef structure using two cameras, an action camera and a compact camera, were compared with its known dimensions detailed in an engineering diagram. According to surface area calculations, the compact camera provided a better 3D mesh than the action camera, with surface area calculations providing an accuracy of 98.2% against the engineering model, compared to 93.2% for the action camera. Images taken at a height of 1 m above the seafloor provided 3D meshes that were more accurate than those using images taken at 2 m above the seafloor. Two image enhancement techniques, histogram equalisation and contrast limited adaptive histogram equalisation (CLAHE), were then applied to assess if this improved the SfM photogrammetry mesh. The 3D mesh from images using the action camera that had a histogram equalisation enhancement provided the most comparable surface area measurement to the engineering diagram, with 100.6% accuracy, indicating our mesh had accounted for growth of benthic organisms on the structure since its installation. In contrast, raw (not enhanced) images had most comparable surface area measurement (98.2% accuracy) using the compact camera. However, the higher apparent accuracy of surface area measurements with the action camera following image enhancement may also be an artefact of inaccurate visual representations from the 3D mesh. Given the comparable accuracy of both approaches, we suggest SfM photogrammetry in turbid benthic environments uses cameras with a larger sensor sizes and customisable settings. This will result in the most accurate 3D meshes from raw imagery, particularly with images taken at a close distance (e.g., [≤] 1 m above the seafloor) and at high intervals (0.5 sec) with percentage overlap (>70%) among images. As the artificial reef in this study was in shallow water (3-4m), lights and/or strobes should be taken into consideration in deeper turbid waters but can also cause problems such as backscatter. Lastly, image enhancement can provide a means to improve image quality, and overall 3D mesh accuracy, when raw image quality and choice of cameras settings were poor.