Hand detection

This is the project for Machine Vision classes at Faculty of Mechatronics, Warsaw University of Technology.

The aim was to write a program which would detect and track a hand.

Required conditions

The program works pretty well provided that following conditions are met:

• The background is relatively static (dynamic background messes everything)
• Works best with bright stray light (for high contrast)
• Only one hand is visible
• All fingers of the hand are visible
• Generally program works better if the face isn't in the image

Algorithm

1. First 500 frames are used to create a good BackgroundSubstractor model
2. Binary Threshold
3. Morphological open
4. Finding contours with findContours() function
5. Finding the contour with greatest area
6. Finding the hulls and defects of the contour
7. Count the rough_palm_center as a mean of defect points, also count standard deviation
8. Remove all defect points which are father than 1,41*std_deviation from rough_palm_center
9. Go back to step 7 until std_dev > Min_rough_center_std_dev
10. Make vector of distances of defects to rough_palm_center
11. Sort the vector
12. Take the three closest defects and count the center and radius of a circle containing them
13. Check if the center didn't move too far away and whether the radius isn't too big
14. If it's ok, count the mean of 10 last palm centers
15. Resulting circle is the center of palm and size of the hand.

Contact

Piotr Pokorski

[email protected]