October 31, 2006

Target Colours

I’ve started coding a mock up of the algorithm using Java (simply because it’s so easy using Java!) and have implemented the portion where the target colour is chosen. The idea is that during the robots startup phase, the person stands centrally to the robot wearing a plain T-Shirt that doesn’t have the same colours as the background. This initial colour is the colour the robot will use as part of its blob recognition to track the person.

Currently the code will take in a picture of someone standing central in the image. Using our models: Chris, Anthony and me, and my wonderous phone camera skills, the program analyses each pixel in the image systematically and turns the pixel white if there’s a hit, and black if it’s a miss. This is done with some basic statistical comparisons but not before every colour is normalised so that is completely discards any information regarding the intensity of the colours. This means that even if the lighting is very very dark, it is still able to know what colour it is. An averaging filter is then applied on the binary image to calculate the position of the person. This will discard any isolated pixels in the resultant image, therefore eliminating noise. Below are three of the pictures I’ve used and the binary image produced after applying the program. The intersection of the green lines indicate the centroid of the main blob, indicating the calculated position of the person.

Ant ProcessedChris ProcessedHoi Processed

- 2 comments by 1 or more people Not publicly viewable

  1. I think there is something wrong with the pictures from your phone, Hoi. The pictures of me and Ant both take something like 25-30 secs to run on my computer. Any other images I try come up almost immediately. I have tried lots of photos from 320×240 to 1000×1500 and they all come up within 2-3 secs max.

    01 Nov 2006, 00:15

  2. No way! That’s really really weird. I guess I can leave the photography to you then :P

    01 Nov 2006, 01:21

Add a comment

You are not allowed to comment on this entry as it has restricted commenting permissions.

October 2006

Mo Tu We Th Fr Sa Su
|  Today  | Nov
2 3 4 5 6 7 8
9 10 11 12 13 14 15
16 17 18 19 20 21 22
23 24 25 26 27 28 29
30 31               

Search this blog



Most recent comments

  • The reflective writing is submitted individually so make sure you submit it yourself tomorrow mornin… by on this entry
  • Yeh Ant's charts look great. Not sure – Ant reckons 1650 max (10% over), so if anyone has any sugges… by on this entry
  • Wow, the report looks really good :) Do you think it will matter too much if it is a couple of hundr… by on this entry
  • nice, overs better than under ;) by on this entry
  • I'll do the other bits then: introduction/overview of person tracking, testing/problems (if Ant+Hoi … by Matt on this entry

Blog archive

Not signed in
Sign in

Powered by BlogBuilder