Research On FPGA Acceleration Technology Of Phase Measurement Profilometry Algorithm

Three-dimensional measurement is a frontier issue of machine vision research,and3 D measurement system is an important part of machine vision system.In order to achieve real-time 3D shape measurement and improve measurement accuracy,many methods have been proposed by researchers.3D reconstruction based on structured light can realize non-contact measurement.At the same time,surface structured light has the advantages of faster reconstruction speed and higher accuracy than linear structured light.The system needs hundreds of thousands or even millions of pixels to traverse when processing an ordinary grayscale image,reconstruction of a scene often needs many such grayscale images,which greatly increases the time required to reconstruct a scene.Secondly,due to the nonlinear response of projector and camera,the phase measurement error leads to the measurement accuracy of 3D reconstruction.Therefore,this paper mainly from two aspects of improving the measurement accuracy and measurement speed.This article is to achieve real-time 3D shape measurement and improve measurement accuracy.Firstly,a three-dimensional measurement algorithm that is conducive to FPGA implementation is selected through algorithm simulation.A high-speed FPGA architecture for phase measurement profilometry(PMP)algorithm is proposed.The hardware part of the system is mainly composed of a digital projector,a high-speed grayscale camera,an FPGA capture card and a PC.The system uses 12 sine fringe patterns to achieve 3D shape measurement,and corrects the non-linear error of the projector to improve the measurement accuracy.In addition,a look-up table method is used to calculate the arctangent function,which not only improves the calculation speed,but also improves the phase calculation accuracy.The system utilizes the parallelism of FPGA and the superior performance of pipeline processing to redesign modules such as data caching,phase wrapping,phase unwrapping,and obtaining point clouds,which greatly improves the speed of 3D reconstruction.A new point cloud extraction algorithm is proposed,which improves the shortcomings of polynomial fitting algorithm that consumes a lot of resources,and finally realizes the whole process of 3D measurement on FPGA with smaller resources.The experimental results show that the system can reconstruct complex and discontinuous objects,and the system has a great improvement in reconstruction speed and accuracy compared with traditional PCs.The system proposed in this article is more than 80 times faster than matlab,and nearly twice as fast as multi-threaded CPU.