Introduction
    This page contains software and instructions for Aligned Cluster Analysis (ACA) [1] and its hierarchical version (HACA) [2]. All the functions have been written and documented in Matlab format. We additionally provide C++ implementations of some dynamic programming routines which involve many loops and are notoriously slow in Matlab.
        [1] F. Zhou, F. de la Torre, and J. K. Hodgins, "Aligned cluster analysis for temporal segmentation of human motion," in International Conference on Automatic Face and Gesture Recognition (FG), 2008.
        [2] F. Zhou, F. de la Torre, and J. K. Hodgins, "Hierarchical Aligned cluster analysis for temporal segmentation of human motion," submitted to IEEE Transactions on Pattern Analysis & Machine Intelligence (PAMI), 2010.



Installation
    1. unzip aca.zip to your folder;
    2. Run make.m to compile all C++ files;
    3. Run addPath.m to add sub-directories into the path of Matlab.
    4. Run demoXXX file.



Instructions
    The package of aca.zip contains three folders, two setup files and three demo files. ./data: This folder contains motion capture data.
    ./src: This folder contains the main implmentation of ACA and HACA.
    ./lib: This folder contains some necessary library functions.
    ./make.m: Matlab makefile for C++ code.
    ./addPath.m: Adds the sub-directories into the path of Matlab.
    ./demoToy.m: Segmentation of a synthetic sequence by ACA.
    ./demoToyH.m: Segmentation of a synthetic sequence by HACA.
    ./demoMocap.m: Segmentation of motion capture sequence by ACA and HACA. By using this function, you can obtain results similar to those shown here.



Other Tips
    For each C++ code, we provide its corresponding Matlab version. For instance, you can use "acaFordSlow.m" instead of "acaFord.cpp". They have the same interface in both input and output. The C++ code is faster to obtain result while the Matlab version is easier to understand and debug.
    For the DTAK algorithm and ACA algorithm, we also provide a version in which each searching step in dynamic programming can be locally constrained. Although we didn't use this feature in our paper, we found it is useful to obtain a robust alignment and to speedup the algorithm. Please refer to [3] [4] for more details on constraints in Dynamic Time Warping (DTW).
        [3] S. Salvador and P. Chan, "Toward accurate Dynamic Time Warping in linear time and space," in Intelligent Data Analysis, 2007, pp. 561-580.
        [4] L. Rabiner and B.-H. Juang, "Fundamentals of speech recognition," Prentice Hall, 1993.




Copyright
    This software is free for use in research projects. If you publish results obtained using this software, please use this citation.
@inproceedings{Zhou_2008_6155,
   author    = {Feng Zhou and Fernando De la Torre and Jessica K. Hodgins},
   title     = {Aligned Cluster Analysis for Temporal Segmentation of Human Motion},
   booktitle = {IEEE Conference on Automatic Face and Gestures Recognition (FG)},
   month     = {September},
   year      = {2008},
} 
If you have any question, please feel free to contact Feng Zhou ([email protected]).