基于GPU的散斑三维重建系统

Speckle Projection Systems Based on GPU

散斑相关算法可以用来估计场景的深度信息,但因易受到噪声干扰且计算量大而难以应用在基于普通计算机的三维重建系统中。采取零均值归一化互相关函数(ZNCC)作为相关算法的匹配代价函数,对传统的ZNCC快速计算方法进行修改并将其应用于计算机的通用图形处理器(GPU),实现了实时的场景三维重建效果。对比实验表明,在精度一致的前提下,提出的GPU计算方法的速度是CPU算法的39倍。

Speckle correlation algorithms can be used to estimate the depth information of the scene. However, such methods are easy to be disturbed by noises and inherent with high computational cost. This paper presented a 3D-reconstruction system based on structured light implemented on GPU. Depth was estimated from the correlation of speckle projection image. Zero-mean normalized cross correlation（ZNCC） was adopted as the correlation function. Traditional fast-ZNCC calculation method was modified to implement on GPU platforms. Experimental results show a 39x speed-up is achieved using GPU compared with the same computation cost implemented on CPU platforms.