基于体的可视外壳重建不需要复杂的几何计算，易于实现，但其精度和效率仍不理想。提出一种基于统一计算架构(CUDA)和行进立方体(MC)的体素可视外壳并行优化建模方法，将体素相交判断和等值面抽取过程并行分解，利用CUDA的内核多线程机制加速建模；在MC网格化阶段，提取准确交点并消除冗余边计算；利用像素着色器实现混合权重的纹理映射，提高模型精度。在多摄像机采集装置DreamWorld上的实验结果表明，本文方法能够对多个动态物体进行实时准确的3维建模。

Volumetric visual hull reconstruction is easy to be implemented without complicated geometrical computation, but its accuracy and efficiency are still unsatisfactory. A parallel optimized visual hull modeling method is proposed based on CUDA and marching cubes. It parallelizes the voxel intersection judgment procedure and the isosurface extraction procedure, and implements both of them using multiple threads of CUDA to accelerate the reconstruction. During marching cubes process, exact intersection points are calculated and redundant computation of edges is eliminated. The pixel shader is used to carry out the hybrid weight texture mapping so as to enhance the quality of the reconstructed mesh model. Experimental results on the multi-camera platform DreamWorld have proved that this method can reconstruct multiple dynamic objects accurately in real time.