微型实时多目立体视觉机设计与实现

A Miniature Real-Time Stereo Vision Machine with Multiple Cameras

本文给出一种实时计算场景稠密深度图的多目立体视觉机设计与实现方法 .立体视觉机使用多个微型摄像机同步获取场景图像 ,采用图像修正、LoG滤波、多立体图像对匹配和稠密深度图等并行算法 ,利用FPGA的大规模并行处理能力和各算法间的多级流水线关系 ,在一片FPGA芯片上实现了立体视觉信息的实时处理 .设计的立体视觉机体积小 ,运行速度快 .当图像分辨率为 32 0× 2 4 0像素 ,深度搜索范围为 6 4级 ,深度图精度为 8位 ,时钟频率为6 0MHz时 ,恢复稠密深度图的速度大于 30帧 /秒 .

A miniature stereo vision machine to generate high resolution dense depth maps has been developed for application to portable intelligent robots and smart visual interface.The machine uses multiple cameras,each with a very wide field of view,to synchronously capture stereo image sequences,and then computes dense depth maps in real time. The whole algorithm,including radial distortion correction,LoG filtering,correspondence finding,and dense depth map computation,is compactly implemented in a single FPGA.The machine,running at 60Mhz,could achieve more than 30 fps for building dense disparity maps with 320×240 pixels in 64 pixel disparity search range and 8 bit depth precision.