3D Scene Reconstruction of Vision Information for Mobile Robot Applications

Abstract

Three dimensional scene reconstruction from stereo image pairs plays an important role in different engineering applications such as computer vision, computer graphics, medical diagnosis, animation and more particularly in mobile robot mapping. For 3D scene reconstruction triangulation method is considered as an efficient tool which relies on camera intrinsic/extrinsic calibration parameters and a disparity map. In practice, the image points in the left and right image plane cannot be measured with accuracy due to geometric distortion (for example lens distortion), dispersed intensity, etc. However, the projection rays do not intersect in 3D space and also may not intersect with the point in the 3D space. In this study, an optimised mid-point triangulation method is proposed to reduce an error in the reconstructed data. An iterative refinement process is introduced in this approach to improve accuracy in the depth measurement of a stereoscopic system. The performance of the method is validated by extensive experimentation and the obtained 3D reconstruction data demonstrates better accuracy than the conventional methods.