The alignment distance on Spaces of Linear Dynamical Systems

We introduce a family of group action induced distances on spaces of Linear Dynamical Systems (LDSs) of fixed size and order. The distance between two LDSs is computed by finding the change of basis that best aligns the state-space realizations of the two LDSs, hence the name alignment distance. This distance can be computed efficiently, hence it is particularly suitable for applications in modern dynamic data analysis (e.g., video sequence classification and clustering), where a large number of high-dimensional LDSs may need to be compared. Based on the alignment distance, we also define a notion of average between LDSs of the same size and order with the property that the order and in some cases stability are naturally preserved. Various extensions to the basic notion of alignment distance are also proposed.

[1]  L. Silverman,et al.  Model reduction via balanced state space representations , 1982 .

[2]  W. D. Ray.  Identifiability, Recursive Identification and Spaces of Linear Dynamical Systems , 1991 .

[3]  René Vidal,et al.  Surgical Gesture Classification from Video Data , 2012, MICCAI.

[4]  B. Moore.  Principal component analysis in linear systems: Controllability, observability, and model reduction , 1981 .

[5]  U. Helmke,et al.  Optimization and Dynamical Systems , 1994, Proceedings of the IEEE.

[6]  Stefano Soatto,et al.  Dynamic Textures , 2003, International Journal of Computer Vision.

[7]  R. Vidal,et al.  Histograms of oriented optical flow and Binet-Cauchy kernels on nonlinear dynamical systems for the recognition of human actions , 2009, 2009 IEEE Conference on Computer Vision and Pattern Recognition.

[8]  René Vidal,et al.  Group action induced distances for averaging and clustering Linear Dynamical Systems with applications to the analysis of dynamic scenes , 2012, 2012 IEEE Conference on Computer Vision and Pattern Recognition.

[9]  René Vidal,et al.  Fast Jacobi-type algorithm for computing distances between linear dynamical systems , 2013, 2013 European Control Conference (ECC).

[10]  Roger W. Brockett,et al.  A scaling theory for linear systems , 1980 .

[11]  A. El-Sakkary,et al.  The gap metric: Robustness of stabilization of feedback systems , 1985 .

[12]  David Forbes Delchamps.  The geometry of spaces of linear systems with an application to the identification problem , 1982 .

[13]  U. Helmke.  Balanced realizations for linear systems: a variational approach , 1993 .

[14]  J. Willems,et al.  Parametrizations of linear dynamical systems: Canonical forms and identifiability , 1974 .

[15]  K. Glover.  All optimal Hankel-norm approximations of linear multivariable systems and their L, ∞ -error bounds† , 1984 .

[16]  John M. Lee.  Introduction to Smooth Manifolds , 2002 .

[17]  Michiel Hazewinkel.  Moduli and canonical forms for linear dynamical systems II: The topological case , 2005, Mathematical systems theory.

[18]  L. Younes.  Shapes and Diffeomorphisms , 2010 .

[19]  Stefano Soatto,et al.  Recognition of human gaits , 2001, Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001.

[20]  Dietmar Bauer,et al.  Balanced canonical forms for system identification , 1997, Proceedings of the 36th IEEE Conference on Decision and Control.

[21]  René Vidal,et al.  Group Action Induced Distances on Spaces of High-Dimensional Linear Stochastic Processes , 2013, GSI.

[22]  K. Nomizu,et al.  Foundations of Differential Geometry, Volume I. , 1965 .

[23]  Bernard Hanzon,et al.  Riemannian metrics on spaces of stable linear systems, with applications to identification , 1982, 1982 21st IEEE Conference on Decision and Control.

[24]  J. M. C. Clark,et al.  The consistent selection of parameterizations in system identification , 1976 .

[25]  Wilson J. Rugh,et al.  Stability preserving interpolation methods for the synthesis of gain scheduled controllers , 2000, Autom..