Motion Icon: Stage 2
1. Overview

In the Stage 1, I have proposed and implemented a framework for creating motion icon. This framework can be improved as follows:

(1) Improve the result presentation, including using high-contrast colors for background scenes and the human models, and using visually appealing human model.

(2) Determine camera setting considering the root trajectory instead of the simple major motion axis.

(3) Better pose clustering algorithm is desired. Current greedy algorithm is heavily dependent on the order that frames are processed. Also better distance metrics between the frame and the cluster is desired.

(4) Improve efficiency by reducing dimension in each frame.

(5) Extract key frames through cluster graph analysis.

(6) Better way to re-position the key frames. 

(7) Add more motion expressions, such as action lines and motion blur, to convey motion.

2. Goal at Stage 2

The goal for Project 2 is to improve the current framework, specifically in the following way:
(1) Improve the result presentation, including eliminating alias in the scence, and using high-contrast colors for background and foreground.

(2) Determine camera setting considering the root trajectory instead of the simple major motion axis.

(3) Better pose clustering algorithm is desired. Current greedy algorithm is heavily dependent on the order that frames are processed. Also better distance metrics between the frame and the cluster is desired.

(4) Improve efficiency by reducing dimension in each frame.

 
3. References
[1]James E Cutting. Representing motion in a static image: constraints and parallels in art, science, and popular culture. 
Perception. Vol. 31, 2002: 1165-1193.
[2] Assa, J., Caspi, Y., and Cohen-Or, D. Action synopsis: pose selection and illustration. ACM Trans. Graph. 24 (3), 2005: 667-676.
[3] Feng Liu , Yueting Zhuang , Fei Wu , Yunhe Pan, 3D motion retrieval with motion index tree, Computer Vision and Image Understanding, 
Vol.92 (2-3), 2003: 265-284. 
[4] Sam T. Roweis and Lawrence K. Saul. Nonlinear Dimensionality Reduction by Locally Linear Embedding. Science. Vol. 290, 2000: 2323-2326.
[5] Bongwon Suh, Haibin Ling, Benjamin B. Bederson, and David W. Jacobs. Automatic thumbnail cropping and its effectiveness. 
In Proceedings UIST'03, pages 95-104, 2003.
[6] Feng Liu and Michael Gleicher. Automatic Image Retargeting with Fisheye-View Warping. ACM UIST 2005, Seattle, USA, October 2005. pp.153-162.
[7] Liu, G., Zhang, J., Wang, W., and McMillan, L. 2005. A system for analyzing and indexing human-motion databases. SIGMOD '05. 
ACM Press, New York, NY, 924-926.