Research Topics

Research Projects

Camera Calibration

Modern photogrammetric methods like image mosaicking or 3D reconstruction allow to use images taken by optical cameras to accurately measure distances or volumes of objects in a scene of interest. In order to archive robust results, it is important to calibrate the geometric properties of the camera used. Technically, the corresponding 3D ray in space is determined for each pixel in the image. The DeepSea Monitoring group is interested in developing and experimenting with methods for refractive, underwater camera calibration. These approaches explicitly model light refraction at the underwater housing port, which is important for avoiding a systematic, geometric modeling error in later measurements. Based on such a calibration, it is possible to apply methods like refractive 3D reconstruction to images captured at the seafloor and compute a virtual, digital 3D model of the scene, that can be viewed interactively.For more details, procedures and tools on camera calibration, please visit our camera calibration page.

Optical Bubble Stream Characterization

Size, number, rising speed and release frequency of bubbles are essential parameters for direct flux measurements of free gas release at the seafloor and the dissolution behaviour of gas bubbles in the water column. Such parameters are needed as input for hydroacoustic flux estimates and calculations for methane transport towards the sea surface. In this research, novel methods for visual bubble stream characterization are developed using a wide baseline stereo camera system that photographs rising bubbles at high frame rates in situ (the BubbleBox).

Visual Robotics

Because of the ever-changing requirements and missions at a research institute, GEOMAR needs flexible platforms and vehicles that are versatile, re-configurable and can be programmed to the respective needs. Additionally, such platforms should be controlled ideally on an abstract level. Desirable missions could be "to map a particular area and to come back with a map that contains no holes", "to explore some track and generate a summary of the different seafloor types", "to detect changes in a habitat by repeated scanning" or "to make detailed measurements once something unusual can be seen".For all these scenarios, the robot must process data onboard (real-time machine vision) and adapt the mission when needed.