Video-rate Hair Tracking System Using Kinec (For CCIW2013 Suzuki, Wu, Chen)

a) Experiments for testing the influence of existence of background objects having the color similar to the target, we compared the performance of our method with K-means tracker for the same video sequence.
b) The tracking result of multi-target
   b-1) The tracking result of long black hair before black background
   b-2) The tracking result of long brown hair
c) Some comparative experiments of tracking a face and tracking a head.
d) Some tracking results show in 3D space: head3D.mp4 and hair3D.mp4

Tracking Iris Contour with Variable Circular Sector Separability Filter (For CCIW2013 Wu, Matumoto, Chen)

a) Comparative experiments about accuracy and speed of iris tracking using CCSF and our VCSSF.
b) We applied VCSSF to many image sequences under various person and conditions .
Some results of our EK-means Tracker and Comparative experiment results between K-means tracker

1) Tracking hair using EK-means tracker
2) Tracking hair using K-means tracker The background has similar color (black) but different depth to the target (hair)
K-means tracker failed to update the target ellipse.
EK-means tracker worked correctly
The size of left image was 640x480 pixel, and the right one was 1024x768 pixels.
K-means tracker failed to adjust the target ellipse when the target size in left image bigger than about
300x300 pixels.
EK-means tracker could update the target ellipse correctly for all frames, the biggest size was about 700x700 pixels.
EK-means tracker has no limit for object size in input image.
4) Other tracking results using EK-means tracker for non-rigid object.


Comparative experiment results between
a) Our K-means trackerSANYO-Xacti
b) Template matching with SAD
c) Mean-shift tracking algorithm
Table of the experimental results
The movies in the table are arranged as follows:
Upper-Left: Original sequence Upper-Right: Mean-shift
Lower-Left: Template matching with SAD Lower-Right: Our K-means tracker
(1) (2) (3) (4) (5) (6) (7) (8)