Description
1) Using the images (aerobic-[001-022].bmp) provided on the class WWW site,
experiment with simple motion detection between consecutive frames using (abs)
image differencing. Clean-up and remove any tiny regions (e.g., use techniques such
as bwareaopen, median filtering, etc.). Experiment with different thresholds. [2 pts]
2) Compute an MEI and MHI on the image sequence (using your best motion differencing
approach from problem #1 above for each image pair i and i-1), simulating the current
MHI “timestamp” for each image pair using the larger of the image pair index values
(i.e., use i, not i-1). You will have difference images from i=2 to 22. The MEI/MHI
duration should include all image diff results in the sequence into the final template.
Use imagesc (Matlab) to show your results. Compute the 7 similitude moments for the
final MEI and the MHI (make sure to normalize the MEI and MHI values to be
between 0-1 before computing the moments using the given formula in the class notes:
max[0, (i-1.0)/21.0] for this example). [4 pts]
3) Create a 101×101 image with a black (0) background and a white (255) box of size
21×21, placing the upper-left corner at pixel (row=40, col=6). Create another new box
image, but shift the box 1-pixel to the right and 1-pixel down. Compute the normal
flow between the images. Use MATLAB’s quiver function to draw the vector
motions on the image (call imagesc, then ‘hold on’, and lastly call quiver). (Make
sure your gradient mask orientations/directions and the plot axes are consistent!!!)
Make sure all masks are “correct” with proper scaling/normalization. Is the result
what you expected? Why or why not? Comment on the flow for the 4 sides of the box
and also the 4 corners. [5 pts]
4) As usual, turn in all code, printouts of images, and discussion of results. Make a script
(with the proper naming) to do the above tasks and call needed functions. Upload
your code and images to Carmen.