The Visual Recognition Of Shadows By An Active Observer
Gareth-David Funka-Lea · 1994
In computer vision for object recognition or autonomous navigation, shadows are a frequent occurrence. However, shadows in an image can make it difficult to partition the image into regions corresponding to physical objects. Consequently, shadows must be accounted for in images. Despite this, relatively little work in image understanding has addressed the problem of recognizing shadows. This is in large part because shadows are difficult to identify. They cannot be infallibly recognized until a scene's geometry and lighting are known. However, this dissertation present a number of cues which together strongly suggest the identification of a shadow and which can be examined without a high computational cost. The techniques developed are: a color model for shadows and a color image segmentation method that recovers single material surfaces as single image regions irregardless of whether the surface is partially in shadow; a method to recover the penumbra and umbra of a shadow; a method for determining whether some object could be obstructing a light source; and a set of tests to determine that shadows that appear to be cast on planar surfaces are in fact on those surfaces. These cues address both the spectral and geometric nature of shadows. Although some cues require the recovery of information about scene geometry, this dissertation shows how these cues can be applied without doing absolute depth recovery for surfaces or objects in the scene--with the exception of an estimate of the light source position, which is part of the obstruction cue. These cues either depend on or their reliability improves with the examination of some well understood shadows in a scene. Consequently, the observer is equipped with an extendable probe for casting its own shadows. Any visible shadows cast by the probe are easily identified because they will he new to the scene. These actively obtained shadows allow the observer to experimentally determine the number, location, and rough extent of the light sources in the scene and the observer gains information about the likely spectral changes due to shadows. The system has been tested against a variety of indoor and outdoor scenes.