next up previous contents
Next: Depth Judgements and Up: Computer Vision Previous: Geometric Model Construction

Monocular Stereopsis: Passive Range Imaging with a Multiple Aperture Camera

We have developed a technique for passively sensing the three-dimensional structure of a scene using a single camera. The iris of a conventional camera is replaced by a mask with multiple apertures, forming a composite image that is analyzed to compute depth. Unlike binocular stereopsis, the views from each aperture are superimposed, so that conventional methods in stereo vision do not apply. Still, the local displacement between corresponding points in these views is related to their distance from the camera. This depth cue provides the basis for a new paradigm in passive range sensing - monocular stereopsis. Our technique is based on modelling the formation of the composite image as an echo process, where the depth of a point in the scene is directly related to the spatial delay of its visual echo. Cepstral analysis is the method used to detect this echo, and our model of the composite image cepstrum allows measurement of monocular disparity to sub-pixel precision, as well as an estimate of its associated error distribution. This data, computed over a dense grid, is interpreted to construct a piecewise planar representation of surfaces in the scene, based on a maximum likelihood criterion. Borrowing techniques from visual psychophysics, the spatial resolution of our range sensor has been evaluated in terms of an intelligent agent making decisions about its environment. This new range imaging technique has been successfully applied to real-world scenes, demonstrating that it is well-suited for mobile robot navigation and obstacle avoidance.

D.G. Jones, D. Lamb

  [{figure}centersubfigure[]epsfxsize=3inleavevmodeepsfboxfigures/lounge_overlay.pssubfigure[]epsfxsize=3inleavevmodeepsfboxfiguresview2_disp.pssubfigure[]epsfxsize=3inleavevmodeepsfboxfiguresview2_img.pslabeldjones:figure3center{figure}]
Figure: Monocular Stereopsis for Mobile Robot Navigation. (a) Composite image of an indoor scene, taken with a double aperture camera. (b) Raw monocular disparity map, obtained by cepstral analysis of the composite image. (c) Range data computed by monocular stereopsis from five different viewpoints, superimposed onto on an actual map of the room. The five viewpoints correspond to points along the path of a mobile robot; the arrows and dotted lines indicate the line of sight and effective field of view, respectively.



next up previous contents
Next: Depth Judgements and Up: Computer Vision Previous: Geometric Model Construction



Thierry Baron
Mon Nov 13 10:43:02 EST 1995