Environment modelling using spherical stereo imaging

We propose an environment modelling method using high-resolution spherical stereo colour imaging. We capture indoor or outdoor scenes with line scanning by a rotating spherical camera and recover depth information from a stereo image pair using correspondence matching and spherical cross-slits stereo geometry. The existing single spherical imaging technique is extended to stereo geometry and a hierarchical PDE-based sub-pixel disparity estimation method for large images is proposed. The estimated floating-point disparity fields are used for generating an accurate and smooth depth. Finally, the 3D environments were reconstructed using triangular meshes from the depth field. Through experiments, we evaluate the accuracy of reconstruction against ground-truth and analyze the behaviour of errors for spherical stereo imaging.