Learning Switching Linear Models of Human Motion

By Vladimir I. Pavlovic, James M. Rehg, John MacCormick. Neural Information Processing Systems, 2000

436 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/c8267434d8c9b188fc9f39f8a451b28b98d2233d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Layered Dynamic Textures (2005, methodology): “With respect to time-series models, the LDT is related to switching linear dynamical models, which are LDSs that can switch between different parameter sets over time [33], [34], [35], [36], [ 37 ], [38], [39], [40].”
Explicit-Duration Markov Switching Models (2014, background): “The SLGSSM enables sophisticated modelling and estimation of hidden dynamics underlying noisy observations [Pavlovic et al., 2001, Zoeter, 2005, Mesot and Barber, 2007, Chiappa, 2008, Quinn et al., 2009].”
HP-GAN: Probabilistic 3D Human Motion Prediction via GAN (2017, methodology): “Most work in probabilistic human motion prediction uses non-deep learning approaches [28, 33, 38, 39, 20, 21, 7].”
On Human Motion Prediction Using Recurrent Neural Networks (2017, background): “Traditional approaches have typically imposed expert knowledge about motion in their systems in the form of Markovian assumptions [25, 32], smoothness, or low dimensional embeddings [48].”
Prior Knowledge Improves Decoding of Finger Flexion from Electrocorticographic Signals (2011, background): “) Introduces switch states that form a switched linear dynamic system (SLDS).”
Machine Recognition of Human Activities: A Survey (2008, methodology): “Pavlovic and Rehg [89], [90] model the nonlinearity in human motion in a similar framework, where the dynamics are modeled using LDS and the switching process is modeled using a probabilistic finite state machine.”
Probability Product Kernels (2004, methodology): “This is true for the switching linear dynamical system (SLDS) (Pavlovic et al., 2000), which combines the discrete hidden states of the HMM and the continuous state space of the LDS:”
Neural Granger Causality (2021, methodology): “Human motion has been previously modeled using both linear dynamical systems [62], switching linear dynamical systems [37], [63] and also nonlinear dynamical models using Gaussian processes [64].”
Convolutional Sequence to Sequence Model for Human Dynamics (2018, background): “Previously, Hidden Markov Model [2], linear dynamics system [15], Gaussian Process latent variable models [19] etc.”
Sharing Features among Dynamical Systems with Beta Processes (2009, methodology): “6 Motion Capture Experiments The linear dynamical system is a common model for describing simple human motion [11], and the more complicated SLDS has been successfully applied to the p roblem of human motion synthesis, classification, and visual tracking [17, 18].”
3D People Tracking with Gaussian Process Dynamical Models (2006, background): “Switching LDS and hybrid dynamics provide much richer classes of temporal behaviors [8, 14, 15].”
Coupled Action Recognition and Pose Estimation from Multiple Views (2012, background): “2007, 2010) or switching models (Pavlovic et al. 2000; Jaeggli et al. 2009; Chen et al. 2009) that rely on action-specific manifolds have been shown to be more flexible.”
A review of change point detection methods (2018, background): “They are widely used in certain areas such as speech recognition [116], brain imaging [114], video segmentation [109] and bioinformatics [21, 44, 48].”
Two Distributed-State Models For Generating High-Dimensional Time Series (2011, methodology): “Furthermore, the resulting motions often grossly violate dynamics, resulting in artifacts such as “footskate” and thereby requiring extensive clean-up using inverse kinematics.”
Bayesian Nonparametric Inference of Switching Dynamic Linear Models (2010, background): “INTRODUCTION Linear dynamical systems (LDSs) are useful in describing dy namical phenomena as diverse as human motion [3], [4], financial time-series [5]–[7], maneuverin g targets [8], [9], and the dance of honey bees [10].”
Hierarchical Aligned Cluster Analysis for Temporal Clustering of Human Motion (2013, methodology): “Because exact inference in SLDS is intractable, Pavlovi (cid:2) c et al. [35] proposed approximate inference algorithms by casting the SLDS model as a dynamic Bayesian network (DBN).”
Recurrent Network Models for Human Dynamics (2015, methodology): “Prior work has tackled synthesis of stylized human motion using bilinear spatiotemporal basis models [1], Hidden Markov Models [3], linear dynamical systems [21], and Gaussian process latent variable models [46, 40], as well as multilinear variants thereof [12, 45].”
Gaussian Process Dynamical Models (2005, background): “, [11]), and nonlinear transition functions, such as represented by Radial Basis Functions [2].”