Ground Plane Obstacle Detection under variable Camera Geometry Using a Predictive Stereo Matcher
A scheme is proposed for ground plane obstacle detection under conditions of variable camera geometry. It uses a predictive stereo matcher implemented in the PILUT architecture described below, in which is encoded the disparity map of the ground plane for the different viewing positions required to scan the work space. The research is the extension of Mallot et al’s (1989) scheme for ground plane obstacle detection which begins with an inverse perspective mapping of the left and right images that transforms the image locations of all points arising from the ground plane so that they have zero disparity: simple differencing of the resulting images then permits ready detection of obstacles. The essence of this physiologically-inspired method is to exploit knowledge of the prevailing camera geometry (to find epipolar lines) and the expectation of a ground plane (to predict the locations along epipolars of corresponding left/right image points of features arising from the ground plane).
Unable to display preview. Download preview PDF.
- S.H. Lane, M.G. Flax, D.A. Handelman, J.J. Gelfand, “Function approximation using multi-layered neural networks with B-spline receptive field functions.”, CSL Report 47,1991, 1–37Google Scholar
- H.A. Mallot, E. Schulze, & K. Storjohann, “Neural network strategies for robot navigation.”, Proc. n’Euro,In G. Dreyfus & L. Personnaz (Ed.), 1988 ,Paris:Google Scholar
- J.E.W. Mayhew, P. Dean,P. Langdon, “Artifical neural networks for the kinematic control of a stereo camera head”, (in preparation), 1992Google Scholar
- T. Poggio,F. Girosi, “A theory of networks for approximation and learning.”,A.I. MEMO NO. 1140.,Atifical Inteligence Laboratory, Massachusetts Institute of Technology, 1989Google Scholar