ECE 5760: Final Project

ROS Stereo Vision Accelerator

Jay Fetter(

Tian Yao (

Raghava Kumar (

Introduction    top

Full setup
Demo of Stereo Vision

A stereo camera is a camera that, through the use of two or more lenses, can capture three-dimensional images. Our project turns the FPGA into a device that is able to reduce the time it takes to generate stereo images from rectified image sources. We then integrated our system into a ROS node, which allows our hardware to be easily used by any ROS system.

Our node subscribes to rectified images and publishes stereo images. We process the input images by passing them through a convolution filter followed by a sparse census transform. We then pass the output of the census transform to a census windowing module. This procedure preprocesses the image which is then passed into the correspondence module to finally generate the stereo images.

We were successful in our project and implementing the stereo image computation algorithms. We were able to accelerate this process from a 1 minute and 30 second generation time to a few milliseconds on our hardware.