Multi-ocular 3D vision by back projection method

We have been researched the 3D object measurement using a back projection method. Basic idea of this method is that all the intersection points of back projection lines (e.g. sight lines) which pass through the lens center of the camera and feature points on the scanned images are chosen as candidates of the 3D position of the features, and then the locations where such intersections gather are considered as the real locations of the feature points in the 3D space. A 3D direct voting method to find voxels which large number of sight lines pass through is also reported. In these methods, a major problem is how to remove or to reduce the false intersections. In order to solve this problem, we apply the trinocular vision method. By introducing this, we can remove the false intersections effectively to make the 3D measurement possible using rather small number of images.