Fast Stereo 3D Imaging Based on Random Speckle Projection and Its FPGA Implementation

Yuhao Shang, Wei Yin, Shijie Feng, Tianyang Tao, Qian Chen, Chao Zuo

Abstract: In this paper, we propose a fast stereo 3D imaging technique based on random speckle projection and its FPGA implementation. Stereo vision, as a classic passive method for 3D shape measurement based on the multi-view geometric constraints, can realize the 3D reconstruction of the tested scene using a pair of images captured through the binocular cameras. In addition, some complicated matching techniques, such as graph cut and block matching, are used to obtain a global disparity map but it leads to massive computing overhead. To solve this problem, we developed a fast stereo vision system based on FPGA. Benefiting from the full parallel architecture of FPGA, the complete computational framework is based on a full pipeline design, that is, the storage and calculation of data are performed under the system clock to implement different works of stereo vision (including stereo rectify and stereo matching) at the same time, promoting calculation speed and measurement efficiency. In order to further improve the accuracy of 3D measurement, by introducing structured light illumination into the existing system, a projection system based on random speckle is designed where fast speckle projection and synchronous acquisition are realized on the FPGA hardware. Experimental results verify that our method can achieve high-speed and robust 3D shape measurement.