Calibration And Reconstruction Optimization Of Monocular 3D Imaging System Based On DSP

Reconstructing 3D spatial information from 2D image is called digital three-dimensional imaging technology. This technology has been widely used in manufacturing, medical, aerospace and 3D printing and other fields in recent years. However, the traditional 3DI technology is dominantly based on PC. The PC imaging system is inconvenient due to poor flexibility and high power cost. In order to satisfy the increasing production demands and realize online 3D inspection, the research of 3DI system based on embedded platform has been carried out. We develop a calibration technology which is uncomplicated, flexible and can be operated on site with important academic significance and wide applications. This thesis is sponsored by national natural science fund project entitled “High Precision 3D Vision Measurement based on Telecentric Imaging and Generalized S-transform”(E051102) and high-level talents project of Guangdong province project entitled “The Research of Key Techniques for High Precision and High Speed 3DI ”(No. 2011125). This paper researched some key techniques based on embedded DSP platform developed by our research team, such as monocular calibration and 3D reconstruction optimization. The main research contents and achievements include the following aspects: 1. The calibration technologies of monocular vision system are analyzed in detail and a monocular calibration method based on virtual target is proposed. It can accomplish the camera calibration and the projector calibration at the same time without the assistant of additional equipment. Finally optimizing them globally. This method improves the calibration speed and flexible operation for the monocular 3D system and makes the calibration much easier and simpler. The experiment shows that the root mean-square error of reverse projections of projector is less than 0.17 pixels. 2. Accomplish the transplantation and optimization of the phase-aided 3D reconstruction algorithms based on embedded DSP platform. And then under the integrated development environment of DSP, optimize the system with respect of several relative technical skills, such as compiler optimization, loop unrolling, software pipelining, fixed-point operation is used instead of floating-point operation and CORDIC algorithm. The experiment results show that the program execution efficiency was increased by nearly 2.4 times. With a resolution of 640×512 pixels reconstructed in a single time, the total time is about 2s. 3. The monocular vision system based on embedded 3DI platform has been set up, and the system calibration and the optimized 3D reconstruction algorithm have been tested. Finally verified the effectiveness by reconstructing the calibration ball and the inspected object. The experiment results show that this system can realize 3D fast reconstruction with high precision.