Davies, C.J., Nixon, M.S. and Carter, J.N.
Acquisition of Surface Normal Vectors via Coloured Spot Projection.
Full text not available from this repository.
Acquisition of scene features such as depth information and surface orientation is one of the key problems in three dimensional (3D) vision. A common 3D sensing solution is that based upon triangulation, where surface position is perceived from the displacement of a projected point or stripe. To calculate the depth information for a complete scene requires a single projected element to be scanned across the scene. The scanning process makes such data acquisition unsuitable for dynamic scenes where surfaces may be moving. Illuminating the whole scene with a single projected pattern removes the need for scanning and enables 3D data to be captured at video rate. Matching the imaged pattern elements with those projected is essential if triangulation is to be used to calculate surface position. Due to occlusions or surface discontinuities the imaged order of pattern elements may be different to that projected. This spatial discontinuity problem may be resolved in several ways. Geometrical constraints may be applied to simplify the problem of matching the imaged pattern elements with those projected. For a calibrated system the possible locations of the projected pattern elements in the image are known; the epipolar constraint. Hu and Stockman used the epipolar and other constraints to solve the matching problem for a projected grid. A problem is that there can be more than one labelling for the entire grid which satisfies all the constraints. Another approach to resolve spatial discontinuity is to encode the projected pattern. Boyer and Kak developed an experimental system that acquires the entire range map of a scene from a single projected pattern of coloured stripes. Practical systems have been developed. Rather than indexing the projected stripes using colour, Vuylsteke and Oosterlinck suggest a novel binary illumination encoding scheme. A black and white chess board pattern is used, with binary information encoded at the corners where squares intersect. Both of the above encoded schemes are based on projecting one dimensionally encoded stripes. Griffin et al have described a method for generating a two dimensionally encoded pattern and suggested it would be suitable for a pattern of coloured spots as well as monochromatic pattern elements. In this new system, using spots confers isotropy of shape and hence the resolution is the same in either dimension, colour is included to resolve spatial discontinuity. A further advantage is that analysis of the perceived spots can capture position and local orientation as an array of surface normal vectors. This system therefore combines the advantages of a number of earlier approaches.
Actions (login required)