Incremental point-based integration of registered multiple range images

Integration and reconstruction of object surface model from range images captured from different viewpoints is a primitive problem for the application of laser scanning systems in reverse engineering and industrial inspection. Normally all these range images have to be first transformed into a global coordinate system, space divided algorithm is then employed to integrate and reconstruct the object surface model. But the final output cannot in general guarantee to be a single, smooth and water-tight surface due to large registration errors, noisy points and limited resolution of point sampling. In this paper, we propose a practical system for integrating registered range images from different views. The new method not only combines advantages of points based and mesh based integration such as flexibility and incremental property, but also fixes their shortcomings such as shape distortion, thick surface or difficult determination of grid size. It can also process both unstructured and structured points. The experimental results show that the proposed algorithm is effective and accurate for the integration of registered range images from multi-views with relatively large registration errors and heavy scanning noise.