Fitting plane algorithm-based depth correction for Tyzx DeepSea stereoscopic imaging

The work presented in this paper deals with the poor performance of depth image generation given by Tyzx stereo vision system under different lighting conditions in both indoor and outdoor environments. For that aim, we introduce a fitting plane algorithm to correct distance information as well as fulfill the missing points in the original depth. First, the color image is over-segmented into many small homogeneous regions of interest. Those small regions can be approximately considered as planar surfaces which form the 3-d scene. While 3D points inside each small region should found a plane, this insight is then used to enhance the depth image. Assuming that the environment is made up of a number of small planes, we certainly make no explicit assumptions about the structure of the scene; this enables the algorithm to cope up with many different scenes even with significant non-vertical structure. The algorithm has been confirmed to be easily implemented and robust throughout many experiments in different lighting conditions and different scenarios in both indoor and outdoor environments. Concretely, the proposed approach enables a 3-d reconstruction capability using Tyzx DeepSea G3 vision system which is infeasible from the raw depth data. Moreover, the proposed algorithm improves more than 48% of 3-d reconstruction accuracy compared with the original result given by the stereo vision system over testing 611 scenes under real-time constraint.