3-d Reconstruction from an Uncalibrated Structured Light Sensor

With the aim of providing our structured light sensor with adaptive capabilities, an uncalibrated reconstruction method is presented here. It allows us to simultaneously determine the projection matrices and the co-ordinates of 3-D points in a projective frame by only using pixel correspondences. Then, by translating geometrical knowledge about the scene into mathematical constraints, an Euclidean reconstruction (i.e. a reconstruction up to a similarity transformation) can be performed. In addition, we show how the pattern projection can be used to grab geometrical information about the scene. This report contains also a brief recall on projective geometry, which is the basis of the method, and some experimental results realised both on simulated and real data.