Trinocular Stereo Vision Method Based on Mesh Candidates

One of the most interesting goals of machine vision is 3D structure recovery of the scenes. This recovery has many applications, such as object recognition, reverse engineering, automatic cartography, autonomous robot navigation, etc. To meet the demand of measuring the complex prototypes in reverse engineering, a trinocular stereo vision method based on mesh candidates was proposed. After calibration of the cameras, the joint field of view can be defined in the world coordinate system. Mesh grid is established along the coordinate axes, and the mesh nodes are considered as potential depth data of the object surface. By similarity measure of the correspondence pairs which are projected from a certain group of candidates, the depth data can be obtained readily. With mesh nodes optimization, the interval between the neighboring nodes in depth direction could be designed reasonably. The potential ambiguity can be eliminated efficiently in correspondence matching with the constraint of a third camera. The cameras can be treated as two independent pairs, left-right and left-centre. Due to multiple peaks of the correlation values, the binocular method may not satisfy the accuracy of the measurement. Another image pair is involved if the confidence coefficient is less than the preset threshold. The depth is determined by the highest sum of correlation of both camera pairs. The measurement system was simulated using 3DS MAX and Matlab software for reconstructing the surface of the object. The experimental result proved that the trinocular vision system has good performance in depth measurement.