Institut für
Robotik und Prozessinformatik

Deutsch   English

Zur Zeit leider nur in Englisch vorhanden...

Efficient Shape from Bar Pattern Illumination

3d shape recovery is an important field of research since many years and many publications deal with shape reconstruction techniques. Many of these systems are based on triangulation to compute range data. An other class of techniques estimate surface normals instead of absolute range data, such as shape from texture, shape from shading, shape from specularity, as well as our new approaches. An advantage of the second class of systems is, that no correspondence problem has to be solved and if surface properties are needed, that they directly generate surface normals without the necessity to derive them subsequently from noisy range data. Surface normals are an important basis of robust 3d feature determination, as for example relative surface orientations, curvatures, local maxima, etc. of free-form surfaces. Our proposed approaches generates surface normals by analyzing local orientations and widths in one camera shot of a scene illuminated with one bar illumination or analyzing only local orientations in two shots of two bar illumination rotated relative to each other (see Figure1). In the following we will introduce the approach which uses only one bar pattern projection.

Figure 1: System setup with one bar pattern projector and one camera

The surface reconstruction can be subdivided into several functional steps (see Figure 2).

Figure 2: Processing steps of our shape reconstruction technique

First, we take the grey level image of the scene illuminated with a bar pattern. Undesired information, like inhomogeneous object shadings and textures may be eliminated by an optional appropriate preprocessing. Subsequently, local edge directions and widths of bars are measured. This leads to an ‘angle image’ and a ‘width image’. Missing measure values between the bars are augmented by simple linear 1d interpolation. On the basis of the local bar directions and bar widths we calculate the local surface slopes or surface normals. The surface normals can be used to reconstruct the surface itself, or they can be utilized as basis for 3d feature computation.

Figure 3: Reconstruction of a styrofoam head:
(a) camera shot; (b) reconstructed range map; (c) 3d plot of the range map;
(d) corresponding rendered grey level image obtained by virtual illumination

Determination of local bar angles can be performed by well-known gradient operators. High gradient magnitudes can be used to mask reliable edge angles, i.e. eliminating erroneous ones. The masking of valid angles can be realized by using local maxima for each 1d scan line. After elimination of erroneous angle values, they will be replaced by 1d interpolation. Determination of local bar widths should be done on a sub-pixel level in order to achieve a high accuracy. After estimation of local bar orientations and bar widths in 2d, the local surface normals can be computed.
For experimental evaluation we used a of the shelf CCD-camera and a conventional video beamer. It is also possible to use an ordinary slide projector for bar pattern illumination. The following images (Figure 3) show the reconstruction result of a styrofoam head.

It took 0.24s to generate this page.