|
Computer Vision for Artists and Designers
http://www.flong.com/writings/texts/essay_cvad.html
This page offers a basic introduction to the field of Computer Vision and its application potential for artists and designers.
It also contains some source code written in Processing for some very simple, low-level computer vision and image analysis functions.
Processing Forum: Motion Tracking
http://processing.org/discourse/yabb_beta/YaBB.cgi?board=VideoCamera;action=display;num=1117818806
This link leads to a Processing forum dealing directly with motion capture concepts. The discussion is geared directly for the Processing API,
which is useful. There are also some very good links within the discussions, including one to eyesweb.org (listed below) and another to V3ga's
(listed below) blob detection source code.
V3ga Home
http://www.v3ga.net/index.php?type=0
This is the homepage of v3ga (mentioned above), a programmer with open source code examples for his own computer vision algorithms written in Processing.
It includes video samples of his algorithms at work, with a link to the source code right underneath. Some examples of interest include the
Blob Detection and Edge Detection examples.
EyesWeb.org
http://www.eyesweb.org/
Thise site (mentioned above) allows for free downloads of the EyesWeb package. According to the Processing Forum listed above, EyesWeb offers
a fast and free application for finding edge-detection and performing gesture and motion tracking. Open Sound Control can then be used within
Processing to pass in the position, size, and direction vectors that are calculated and output by EyesWeb.
The Code Project: Motion Detection Algorithms
http://www.codeproject.com/cs/media/Motion_Detection.asp
This is a short explanation/example of procedures for motion detection in a continuous video stream. The processes shown approach the problem
by comparing each current frame with the previous frame or a constant background frame, in order to find the difference between the two.
The areas where difference are found are the areas of interest, where motion most likely took place.
Motion Capture: Acclaim's Optical System
Essay Link
This SIGGRAPH essay offers some informative and general background information on Motion Capture. It includes a rationale behind motion capture
systems, a brief history of motion capture, an explanation of current input methods, and more.
The Condensation Algorithm
Condensation Homepage
The site is an all-inclusive resource to the Condensation Algorithm (Conditional Density Propagation). It's purpose is to track agile
moving objects in the presence of dense backgrounds. Example videos are available, which show the power behind this algorithm. Source
code is also available, however I have been receiving errors after downloading these files and attempting to open them on my desktop. It
also appears that this algorithm may incorporate some very complex mathematics, which could prove to be a problem in trying to understand
it, but further research is needed to assess this fact.
Motion Capture by Least Squares
http://e-collection.ethbib.ethz.ch/ecol-pool/inkonf/inkonf_41.pdf
This document provides yet another algorithm analysis designed for motion capture which uses a least squares matching tracking algorithm.
The method extracts 3D information of the shape and movement of the human body using video sequences acquired with 3 CCD cameras. The
discussion is limited to a general explanation of the concepts behind this method, and does not include any source code or algorithmic examples.
Some examples of the algorithm in action show that least squares could be a feasible method for motion capture, however the mathematics behind
it appears complicated and specific.
Motion Capture by Annealed Particle Filtering
http://www.cs.brown.edu/courses/cs296-4/Papers/DeutscherrCVPR2000.pdf
This resource uses a variation of the Condensation algorithm, labeled Annealed Particle Filtering, to receive greater quality results in an
attempt to use less computational complexity. The depicted examples show very high quality results, however, the mathematics again look very
high-level and complex.
Silhouette Based Human Motion Tracking System
http://www.citr.auckland.ac.nz/techreports/2005/CITR-TR-164.pdf
This paper gives a very detailed description of a silhouette-based markerless motion tracking system. The first part explains a process
for automatically generating a human surface model of the upper-torso. This model can then be used during the second part, which explains
the silhouette-based motion tracking process. Again, the process would take a lot of research and help to understand the mathematical
process empowering it.
Markerless Motion Capture from Monocular Videos
http://www.cse.iitb.ac.in/~sharat/papers/motionCapture.pdf
This resource presents a method for finding the 3D spatial locations of joints of the human body from a single camera video sequence.
Various physical and motion constraints are established to incorporate the method. The method is markerless and the explanation is
simplified, so this may prove to be a good starting-point with researching existing methods.
Direct Least Square Fitting of Ellipses (Applet)
http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/PILU1/demo.html
This resource presents a method for approximating a ROI (region of interest) with an ellipse. This can be used to approximate each blob,
which corresponds to a distinct body part, as an ellipse. The endpoints of the ellipses can then be used as joint positions for animating
the digital model. This website has an applet which displays the results of 3 ellipse-fitting methods, and the java source code of the applet
also provided. It also provides very neat MATLAB code for the ellipse-fitting function, which uses least squares to minimize the distance of
the approximated ellipse with the points in the ROI.
Direct Least Square Fitting of Ellipses (Proposal)
http://www.robots.ox.ac.uk/~awf/ellipse/ellipse-pami.pdf
This paper is the formal proposal of the least-square method that is used in the applet that can be found above.
|