Visually guided vergence in a new stereo camera system
Adam Schneider, Namrata Sharma, Bryan Tripp · Journal of Computational Vision and Imaging Systems · 2018
People move their eyes several times each second, to selectivelyanalyze visual information from specific locations. This is impor-tant, because analyzing the whole scene in foveal detail would re-quire a beachball-sized brain and thousands of additional caloriesper day. As artificial vision becomes more sophisticated, it mayface analogous constraints. Anticipating this, we previously devel-oped a robotic head with biologically realistic oculomotor capabil-ities. Here we present a system for accurately orienting the cam-eras toward a three-dimensional point. The robot’s cameras con-verge when looking at something nearby, so each camera shouldideally centre the same visual feature. At the end of a saccade,we combine priors with cross-correlation of the images from eachcamera to iteratively fine-tune their alignment, and we use the ori-entations to set focus distance. This system allows the robot toaccurately view a visual target with both eyes.