基于双目视觉的三维重构算法实现

Implementation of 3D reconstruction algorithm based on binocular vision

基于Altera DE1 SOC开发平台,选用OV7725互补金属氧化物半导体(CMOS)摄像头组进行图像采集,经现场可编程门阵列(FPGA)预处理后由视频图形阵列(VGA)显示器显示。基于双目视觉的三维场景重构算法包含图像预处理、多线程加速立体匹配以及AD-Census匹配,采用了Verilog HDL和统一计算设备架构(CUDA)编程实现。实验结果表明:系统实现了以7帧/s速度计算640像素×480像素大小的深度图;并且可以将以1.8°为间隔的150幅深度图转换为点云,重构出270°左右视野内的三维场景。

A system based on Altera DE1 SOC,which uses the OV7725 complementary metal oxide semiconductors（CMOS） camera group to collect images,and images are displayed by video graphics array（VGA）after field programmable gate array（FPGA） preprocessing. The reconstruction algorithm of binocular vision 3D scene includes image preprocessing,multi-thread acceleration stereo matching and AD-Census matching. The algorithm is programmed with Verilog HDL and compute unified device architecture（CUDA）.Experimental results show that the system calculating a depth map with 640×480 size is achieved at speed of 7 frames per second,and150 depth maps at 1. 8° intervals can be converted into point clouds,3D scenes in about 270° of view is reconstructed.