search

usc-m / SDP-Team-F

SDP Team F - Group 1 (NASA) and Group 16
0 stars 0 forks source link

Normalise pixel values #34

Closed alfredmakes closed 9 years ago

alfredmakes commented 9 years ago

Currently, Brightness completely throws off our object detection (through colour thresholding.) We should be thresholding based on normalized RGB values (and should not have Brightness threshold values)

ShuxinLin commented 9 years ago

I am thinking if we detect the light area, we can minus the value by (255,255,255) which is the RGB of white. But if we do so, the value will be negative which is invalid.

dollym commented 9 years ago

I was thinking on the same line.

ShuxinLin commented 9 years ago

Actually, the camera cannot detect the true color so it seems it's not possible to normalize. One method is to construct object tracking system which is too complex.

ed728 commented 9 years ago

If you vision people don't mind I can try taking a hack at normalizing it.

alfredmakes commented 9 years ago

Feel free. I was going to head back and play with it shortly (sending emails) but go ahead and I'll join you?

alfredmakes commented 9 years ago

Let me know if you work on this @ed728 I had to leave, will be back at 8.30ish to do this if it isn't already. Otherwise will probably just come in tomorrow - no other big vision tasks for the milestone other than this.

ed728 commented 9 years ago

We were doing strategy the last days before milestone, so did not have the chance to touch it. Has this been done yet?

ed728 commented 9 years ago

I think I have found where this needs to be done and how. The videoStream is submitted several video receivers which process the frame, we just need to create a new one and add it in the correct order. Does not seem complicated. Unless you have already implemented at some other stage of the pipeline, assign this to me then.

alfredmakes commented 9 years ago

It's not implemented no, go ahead :) but as always make sure you test it out before pushing, check the fps isn't seriously effected (is it possible for the number of receivers processing the frame to seriously effect performance?)

edit: Just reminding about this, since we already have framerate issues

ed728 commented 9 years ago

Ok, will do it tomorrow morning. (Want to try fixing rotation today.)

ed728 commented 9 years ago

Ok, looping over pixels to normalize or do some other formatting will cut FPS in half. I think we should just perfect calibration. I will still try to integrate some optimizations in the detection code.

alfredmakes commented 9 years ago

Think I have this working with no framerate issue, going to test it more in an hour or so and then push

alfredmakes commented 9 years ago

This works as far as I can tell/test. At least as good as before, essentially.