Stereoscopic measurements using neural networks

Abstract

In stereoscopy, information from two images taken from different viewing points is converted to a 3-dimensional representation. The method through which human beings experience depth in their vision is governed by this principle. The human brain processes images seen by both eyes in order to reconstruct a 3-dimensional image of the outside world. This technique has been used to make 3-D movies. For this purpose, pictures are taken by two cameras separated at a distance of 7-cm (distance between human eyes). Using special techniques during the projection of the movie, the illusion of depth is created.
We have investigated if the technique of stereovision is capable of measuring the size of complex-shaped objects with high accuracy. The long-term aim is to use stereoscopic methods to monitor coral growth. With a maximum growth rate of about 10 cm per year, monitoring coral growth requires a high accuracy of the method used. Therefore we have looked for methods to determine and improve the accuracy of stereoscopic techniques. We believe that this nondestructive technique can be an asset to the traditional techniques used in the Marine Biology Section of the University of San Carlos.