lec1104h - 1 Virtual 3 Virtual 3-D Blackboard: D...

Info iconThis preview shows pages 1–9. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 1 Virtual 3 Virtual 3-D Blackboard: D Blackboard: Andrew Wu REU 1999 awu@uiuc.edu http://www.cs.ucf.edu/~vision (go to REU99) Finger Tracking with a Single Camera Project Goals Project Goals Using computer vision, implement a virtual 3-D blackboard Program will parse 2-D image input, recording corresponding 3-D motion of a users fingertip Motion of users finger will be recognized as a certain type of 3-D gesture 2 Sample Picture Sample Picture Single color camera Static background One person in picture Consistent lighting Skin Detection Skin Detection l Color Predicate Skin-tones in RGB space marked by computer program Trained on several color images with hand-drawn binary masks Color Predicate data structure saved 3 Example training images Example training images Using the Color Predicate Using the Color Predicate l Check RGB values of every pixel in input image l If RGB value satisfies Color Predicate, output as true in output binary image l Median-filter binary output to remove noise and outliers 4 Results of skin detection Results of skin detection Separating regions Separating regions l Next step: demarcate connected skin regions l Simple 8-connectivity algorithm that grows regions l Cull three largest regions (presumably the head and two arms) Three largest regions, pseudo-colored 5 Separating regions Separating regions Find centroids for largest regions (regions alpha blended with original color image for effect) centroids Finding the arm Finding the arm l Assume fastest moving centroid belongs to gesticulating arm l Find largest delta between two skin frames Difference picture between temporally proximal skin images Difference picture for region of largest centroid delta, only 6 Outlining the arm Outlining the arm l Goal: find perimeter of isolated arm segment l Assume contiguous region l If pixel has 4-connectivity with black region, pixel is on the periphery Segmented arm Calculated outline Dot product Dot product Method: find two vectors between the current pixel and the pixels N steps away (N=3, 2, 2 above) Repeat procedure for all pixels in outline, searching for maximum dot product. Idea: largest dot product will be formed by two vectors extending in both directions from the fingertip A dot B > 0 A dot B = 0 A dot B < 0 7 Results of dot product approach Results of dot product approach Very good output Found finger in all cases when given proper outline of arm Perhaps can detect absence of finger from skin outline For test data, found best value of N to be 3 pixels (N being number of pixels to step away for vector calculation) Video Video Program run on 7 continuous frames In sum: finds skin from color, arm from centroid speed, then finger from dot product of pixel outline Tracks finger fairly well 8 Next issue: Occlusion...
View Full Document

This note was uploaded on 06/12/2011 for the course CAP 6411 taught by Professor Shah during the Spring '09 term at University of Central Florida.

Page1 / 29

lec1104h - 1 Virtual 3 Virtual 3-D Blackboard: D...

This preview shows document pages 1 - 9. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online