Three-dimensional reconstruction based on virtual binocular stereo vision with speckle projection

The prism-based virtual binocular stereo vision system (VBSS) offers advantages of compact structure and low cost, but suffers from limitations such as short baseline, narrow field of view and chromatic aberration. This paper presents an approach to improve the 3D measurement accuracy. First, system calibration is optimized by combining multiple spatial constraints with epipolar constraints. Subsequently, a matching strategy employing multi-frame speckle fusion and Census transform is adopted to enhance feature distinctiveness. Thus, robust integer-pixel matching is achieved. Finally, subpixel disparity is obtained via parabolic fitting. Experiments were conducted to investigate the matching precision, thereby determining the optimal number of reliable frames and the inter-frame projection spacing of speckle patterns. The results demonstrate that the combination of five-frame images with 5mm projection spacing yielded optimal reconstruction results. This study provides an effective solution for high-precision 3D measurement using VBSS systems.