Sensor-assisted video mapping of the seafloor

In recent years video surveys have become an increasingly important ground-truthing of acoustic seafloor characterization and benthic habitat mapping studies. However, the ground-truthing and detailed characterization provided by video are still typically done using sparse sample imagery supplemented by physical samples. Combining single video frames in a seamless mosaic can provide a tool by which imagery has significant areal coverage, while at the same time showing small fauna and biological features at mm resolution. The generation of such a mosaic is a challenging task due to height variations of the imaged terrain and decimeter scale knowledge of camera position. This paper discusses the current role of underwater video survey, and the potential for generating consistent, quantitative image maps using video data, accompanied by data that can be measured by auxiliary sensors with sufficient accuracy, such as camera tilt and heading, and their use in automated mosaicking techniques. The camera attitude data also provide the necessary information to support the development of a video collage. The collage provides a quick look at the large spatial scale features in a scene and can be used to pinpoint regions that are likely to yield useful information when rendered into high-resolution mosaics. It is proposed that high quality mosaics can be produced using consumer-grade cameras and low-cost sensors, thereby allowing for the economical scientific video surveys. A case study is presented with the results from benthic habitat mapping and the ground-truthing of seafloor acoustic data using both real underwater imagery and simulations. A computer modeling of the process of video data acquisition (in particular on a non-flat terrain) allows for a better understanding of the main sources of error in mosaic generation and for the choice of near-optimal processing strategies. Various spatial patterns of video survey coverage are compared and it is shown that some patterns have certain advantages in the sense of accumulated error and overall mosaic accuracy.