-
Notifications
You must be signed in to change notification settings - Fork 12
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Capture taking takes a while #7
Comments
While you're at it: it may be more convenient to try detecting lines (with hough transform), find lines which form a rectangle (of cause a skewed one) and use the corners for the homography, so no markers are necessary. Every laser-bed will be kind of rectangular and if the user can specify regions of interest for the corners it may be enough for robust detection. |
"While you're at it... " is quite a pitfall ;) When it works it's easy to improve upon so I would get it working first. It's also important to check how long each step takes. Maybe it issn't even the opencv step in Java. |
Interesting read: |
First speed test.
It outputs the following, the numbers are milliseconds.
So that means:
|
so... next question is how long it takes to apply the homography via a littel C/C++ program.. maybe we can just source this out. Just a litte tool accepting two filenames and 4 coordinate paris via parameters and does the homography... |
We are also having a look at this now.
My first idea is to buffer the markers, the matrix src, the matrix dst, and the holography h. I created a fork here: For the Pi Cam, I am also thinking about writing a tool that utilises the gpu... but not sure if homography is something that can be done with the pi gpu... here is a first pointer: |
Really nice that you guys are looking into this. |
Hi, I am currently working on this as well and I try to move some of the stuff to C++ / GPU computations and to use some Raspberry Pi 2 hardware specific acceleration. @peteruithoven : Which Java version do you use? Just recently I had to reinstall some of the stuff on a Raspberry Pi 2 and there was a notable difference between older and newer Java versions regarding the performance. If I remember correctly, it needed something around 20s on a JDK 6 and it decreased to ~ 8s with a JDK 8. |
FroChr123 is my student... he writes his bachelor thesis that will add QR Codes to VisiCut. They will be used for two things: physical file sharing and augmented reality markers. For the second part, he will need to make VisiCam as fast as possible (real time!). |
The Raspberry Pi 2 / GPU application is located here: This part should nearly be finished. Of course VisiCam needs to be modified a bit to support it correctly, this will be done in the next few days I think. Both applications need to work together to use the speed boost of the hardware acceleration. Basically the image capturing and the very slow image processing are outsourced from VisiCam to visicamRPiGPU. VisiCam: visicamRPiGPU: |
Would it be possible to update the wiki page https://github.com/t-oster/VisiCam/wiki/Raspberry-Pi-installation-on-Raspbian and then close this issue? |
Currently my setup on the Raspberry Pi, using pistill, with image resolution of 720x576 pixels it takes about 33 seconds for the capture to appear in VisiCut after pressing the capture button.
That's one to improve upon in the future.
Moving the heavy stuff to a c++ app might speed things up. I started on a openFrameworks version:
The text was updated successfully, but these errors were encountered: