ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

Noise Image_raw from video stream

asked 2016-12-17 13:50:30 -0500

tn0432 gravatar image

I have 2 cameras looking at an object, estimating the transformation between two camera images using OpenCV feature matching. I have ROS driver for 2 cameras (gscam). When receiving new image topic (/driver/image_raw), image_call_back() function is called to processing the image:

Step 1: CV.bridge to transfer the ROS image to OpenCV image

Step 2: Transfer to grayscale image

Step 3: Remove distortion effect

Step 4: Histogram equalization to eliminate light effect.

Step 5: Feature detection, matching and estimate transformation

When trying with two images, my transformation algorithm is fine and stable. But when I try with the static cameras (2 sequence images streaming and 2 cameras do not move), my transformation algorithm gives unstable result. I guess it is due to the effect of video stream noise. Have you tried it before and do you have any way to eliminate noise in video stream.

Thanks

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2016-12-17 23:37:31 -0500

ahendrix gravatar image

This is a pretty tough question. It's more of a research problem, and not something that can be easily answered online.

There are a few general things that you can do:

  • Find a Scientific journal about machine vision and read articles about depth from stereo methods. Look at their results, how they measure noise, how much noise they have, and what their suspected sources of noise are. If there is a college or community college near you, they probably have at least something about machine vision.
  • Look at OpenCV's stereo depth estimation algorithms and see if they reference any papers or journal articles. Read those.
  • Compare your results to an industry-standard depth estimation system such as OpenCV's . Try to measure noise the same way the benchmarks do.
  • Enroll in a compute vision course, if there's one available nearby
  • Find a local robotics group, join them, and ask around for anyone who may be able to help in person
  • If you're already enrolled in a University, find a local professor and ask if they have time to help
edit flag offensive delete link more

Question Tools

2 followers

Stats

Asked: 2016-12-17 13:50:30 -0500

Seen: 351 times

Last updated: Dec 17 '16