Skip to Main Content
Three dimensional (3-D) vision techniques in the field of Computer Vision aims mainly at reconstructing a scene to find its three dimensional geometrical information. Passive 3-D vision techniques such as computational stereo vision method do surface reconstruction from disparities arising in images of the same scene taken from multiple views. For reconstruction leading to metrical information of the 3-D geometry of the scene the camera pose with respect to some world reference frame and the camera parameters such as the focal length, sensor size has to be accurately known. Such information especially the pose of the camera might not be known in many applications such as in agricultural, under-water explorations since a unique universal frame of reference is not possible. Also the constancy of the internal camera parameters will not be valid in many applications requiring good accuracy in reconstruction. In such cases the cameras used for passive triangulation is said to be un-calibrated. Stereo vision technique generally requires two images and which need not be from two different cameras. The paper is an attempt to use a single moving camera with which the image of the same scene is acquired from two different views. Since the scene geometry and the pose of the camera are unknown the problem to be addressed is close the so called Structure from Motion (SfM) problem in Computer Vision. The reconstruction method developed in the paper is extended further to segment top surfaces of cuboid shaped objects considered as the objects of interest in the scene reconstruction process. Though the information in the case considered is non-metrical the applications that pose such un-calibrated camera as a demand are plenty.
Date of Conference: 3-5 Nov. 2011