Developing a technique for underwater robot vision is a key factor in establishing autonomy in underwater vehicles. A new technique is developed and demonstrated to depict an underwater scene in three dimensions (3D) for use in underwater robot vision. This technique uses passive lighting and the optical properties of water to approximate distances between objects in a scene. Beer's Law is used to describe the change in intensity that light at different wavelengths experiences as it travels through water. If the intensities can be measured, then Beer's Law provides sufficient information to estimate the unknown distance. Applying the equation pixel by pixel to compare two images produces distances from the camera to everything in the image. A shift in color caused by the uneven absorption by each wavelength is similar to comparing the intensities. The color shift produces a relative distance between objects in the scene rather than distances from the camera to the objects. This relative distance may be combined with other techniques to determine the distances from each pixel to the camera.
http://archive.org/details/anewtechniquefor1094555631
Lieutenant Commander, United States Navy
Approved for public release; distribution is unlimited.