Learning latent spatio-temporal compositional model for human action recognition

By Xiaodan Liang, Liang Lin, Liangliang Cao. 2013

39 citations recorded by the source corpus.

Action recognition is an important problem in multimedia understanding. This paper addresses this problem by building an expressive compositional action model. We model one action instance in the video with an ensemble of spatio-temporal compositions: a number of discrete temporal anchor frames, each of which is further decomposed to a layout of deformable parts. In this way, our model can identify a Spatio-Temporal And-Or Graph (STAOG) to represent the latent structure of actions \emph{e.g.} triple jumping, swinging and high jumping. The STAOG model comprises four layers: (i) a batch of leaf-nodes in bottom for detecting various action parts within video patches; (ii) the or-nodes over bottom, i.e. switch variables to activate their children leaf-nodes for structural variability; (iii) the and-nodes within an anchor frame for verifying spatial composition; and (iv) the root-node at top for aggregating scores over temporal anchor frames. Moreover, the contextual interactions are defined between leaf-nodes in both spatial and temporal domains. For model training, we develop a novel weakly supervised learning algorithm which iteratively determines the structural configuration (e.g. the production of leaf-nodes associated with the or-nodes) along with the optimization of multi-layer parameters. By fully exploiting spatio-temporal compositions and interactions, our approach handles well large intra-class action variance (\emph{e.g.} different views, individual appearances, spatio-temporal structures). The experimental results on the challenging databases demonstrate superior performance of our approach over other methods.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

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

In Computer Vision and Pattern Recognition

Citations in the corpus

3D Human Activity Recognition with Reconfigurable Convolutional Neural Networks (2014, methodology): “Second, the structure of our model can be ﬂexibly adjusted during inference, which is a key to improve the capability of modeling complex patterns [41, 22, 14].”
Human action recognition using genetic algorithms and convolutional neural networks (2016, background): “[14] proposed a hierarchical human action recognition system by modeling each observation as”
AOGNets: Deep AND-OR Grammar Networks for Visual Recognition (2017, methodology): “odule [58], the deeply fused nets [63] and the DLA [69]. Grammars. A general framework of image grammar was proposed in [75]. Object detection grammar was the dominant approaches for object detection [10,74,55,41, 39,40], and has recently been integrated with DNNs [60, 61,6]. Probabilistic program i”
A Deep Structured Model with Radius–Margin Bound for 3D Human Activity Recognition (2015, methodology): “First, the conﬁguration of our deep model can be ﬂexi-bly adjusted to adapt to different input videos, and the sig-niﬁcance of this property has been justiﬁed for human action recognition [22, 38, 45].”
AOGNets: Compositional Grammatical Architectures for Deep Learning (2017, methodology): “odule [58], the deeply fused nets [63] and the DLA [69]. Grammars. A general framework of image grammar was proposed in [75]. Object detection grammar was the dominant approaches for object detection [10,74,55,41, 39,40], and has recently been integrated with DNNs [60, 61,6]. Probabilistic program i”
Progressively Parsing Interactional Objects for Fine Grained Action Detection (2016, methodology): “First, joint object detection frameworks are usually based on tree-like graphical models, for example, deformable part-based models [5, 32], And-Or graph models [38, 16], probabilistic graphical models [20, 18].”