Abstract:
This thesis develops a method for estimating motion for object contours in depth
video sequences. First the depth image is thresholded to nd the foreground object. Con-
tours are then extracted from the thresholded image. Given the contours in two frames
my method estimates the motion that transforms the contour in frame 1 to the contour in
frame 2. First the Euclidean Distance Tansform (DT) is computed for the contour in frame
2. The DT and its gradient at each of the points along the contour in frame 1 are used to
estimate the motion between the contours in the frames. I consider the obtained motion an
estimate of the normal
ow between the frames. Finally the normal
ow is used to estimate
a ne motion between the contours. I demonstrate the method on several synthetic images
corresponding to contours of various shapes. Finally I demonstrate the e ectiveness of my
method on several image sequences including household objects, a person raising their leg
and a long sequence of a human walking.