Scalable Multicamera Heterogeneous Processing Architecture for Phase Measuring Profilometry

Abstract

Despite the rapid development of 3D imaging and measurement technology in recent years, achieving high speed and high accuracy 3D measurement is still challenging. Innovative pattern diagram coding technology, flexible and high-precision system calibration methods, and parallel computing hardware platforms are proposed to improve measurement accuracy and speed. The multi-frequency phase shifting method combined with a least-squares based calibration model can provide high-precision 3D reconstruction results at the expense of speed. A heterogeneous processing architecture based on FPGA and CPU is proposed to accelerate the algorithm. From the perspective of optimal computing resource efficiency, we innovatively separate the phase calculation of 16-pattern scheme from point cloud computing, and reasonably allocate software and hardware resources. The hardware resources for a single camera are further multiplexed and expanded into an FPGA structure that can support multi-camera parallel phase calculations. Adding a small amount of resources can get a doubling of performance. Experiments show that each camera in the multi-camera system can independently reconstruct and display images with a resolution of 640 × 480 at a rate of 50.8 frames per second. The fusion using the reconstruction results of different perspectives of multiple cameras can not only make up for occlusion points, but also improve the measurement accuracy. Standard ball measurement experiments show that the standard deviation of the fusion results can reach 0.013mm. The multi-camera heterogeneous processing architecture presented in this paper demonstrates the great applicability to the requirements of high-precision real-time 3D measurement in scientific and engineering applications.

Topics & Concepts