GPU加速三维面形测量

随着通用计算和图形显示需求的不断增加，图形处理器（Graphics Processing Unit，GPU）在医学、科学计算、图像处理等领域得到了广泛的应用。但它在三维测量领域的应用还只是一个开始。文中基于傅里叶变换轮廓术（Fourier Transform Profilometry，FTP）和三频外差法设计了两套三维测量系统，并利用计算统一设备架构（Compute Unified Device Architecture，CUDA）方法，加速了静态或动态物体的三维重建。在三频外差测量系统中，需要利用高速数字投影模块和相机，同步触发采集小视场表面的12个变形条纹图，然后对图像数据进行处理。实验结果表明:对12幅1 360 pixel1 024 pixel大小的图像进行相位展开运算，GPU方法比CPU方法的效率提高了2 089倍。在基于FTP方法的测量系统中，摄像机只需记录一幅变形条纹图，然后拷贝到显存中，并用CUDA编程的算法进行处理，进而重建出物体的三维面形。基于GPU的FTP方法对一幅1 024 pixel1 280 pixel大小的图像进行计算，其计算时间比CPU方法缩短了27倍。

3D shape measurement accelerated by GPU

Driven by the increasing demands of the general purpose in computation and image display, Graphics Processing Unit(GPU) has been developed and used in many fields, such as medical field, scientific calculations, image processing etc.. But, its application in 3D shape measurement is still a beginning. In this paper, two 3D shape measurement systems based on Fourier Transform Profilometry(FTP) and tri-frequency heterodyne method were implemented with Compute Unified Device Architecture(CUDA) technology to speed up their 3D shape construction of a measured static or dynamic object. In the first 3D shape measuring system based on tri-frequency heterodyne method, a high-speed digital projection module and a synchronously triggered camera were used to record 12 deformed fringe images on the surface of a small object. The experimental result demonstrates that the efficiency of the unwrapping phase calculation by GPU is improved 2 089 times than that of CPU for doing same task on 12 images with 1 360 pixel1 024 pixel each. In the second system based on FTP, only one deformed fringe image was recorded by a camera, then transferred into GPU and processed by the programmed CUDA algorithm to restore the corresponding 3D shape. Compared with the traditional processing method by CPU, the time consumption of FTP method completed by GPU is shortened 27 times for a 1 024 pixel1 280 pixel image.