Hi,

I am working on a type of egomotion application which consists of fast
feature corner detection and template matching. Due to the camera that is
in a continuous moving state, the image frames get blurred/ out of
contrast, making it difficult for the corner detection algorithm to reply
with valid corner lists.

I am creating the application on Android, and was hoping that there is
somebody who found a solution for this type of problem. I've already
enabled auto-focus through
Android, and turned on the back LED to compensate for light intensity lost
when pointing the phone to the floor... but still no solution.

I would highly appreciate it if somebody was able to help me with this

--

Search Discussions

  • Kirill Kornyakov at Sep 5, 2012 at 7:22 am
    This question is not Android-specific, but a good research question, so I
    suggest you to go to the answers.opencv.org.

    If camera moves fast and it creates blurry images, there is almost nothing
    to do about that. You can either physically constrain the camera movement,
    or estimate image sharpness and reject blurry images. If the camera grabs
    frames in real-time, you'll have enough sharp frames for your purpose.

    --
  • Lucas Teixeira at Sep 15, 2012 at 4:55 pm
    But how identify the blurry images? The opencv has some function to this
    propose.
    On Wed, Sep 5, 2012 at 4:22 AM, Kirill Kornyakov wrote:

    This question is not Android-specific, but a good research question, so I
    suggest you to go to the answers.opencv.org.

    If camera moves fast and it creates blurry images, there is almost nothing
    to do about that. You can either physically constrain the camera movement,
    or estimate image sharpness and reject blurry images. If the camera grabs
    frames in real-time, you'll have enough sharp frames for your purpose.

    --


    --
  • Kirill Kornyakov at Sep 17, 2012 at 7:52 am
    I know that it is possible, but I don't know how exactly. Some guy said
    that he applied GaussianBlur<http://docs.opencv.org/modules/imgproc/doc/filtering.html?highlight=gaussian#gaussianblur>and then compared to the original image. Difference shouldn't be large if
    the input image is already blurry. But I'm not sure if this approach is the
    best one, try to google for a better one...
    On Saturday, September 15, 2012 8:55:02 PM UTC+4, Lucas Teixeira wrote:

    But how identify the blurry images? The opencv has some function to this
    propose.
    --

Related Discussions

Discussion Navigation
viewthread | post
Discussion Overview
groupandroid-opencv @
categoriesandroid
postedSep 4, '12 at 9:16p
activeSep 17, '12 at 7:52a
posts4
users3
websiteandroid.com

People

Translate

site design / logo © 2018 Grokbase