Trajectons: Action recognition through the motion analysis of tracked features

By Pyry Matikainen, Martial Hebert, Rahul Sukthankar. 2009

285 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Beyond Gaussian Pyramid: Multi-skip Feature Stacking for action recognition (2014, methodology): “Among them, the trajectory based approaches [18, 35, 38, 40, 8], especially the Dense Trajectory method proposed by Wang et al.”
Actions ~ Transformations (2015, background): “More recently, trajectory based approaches [51, 31, 29, 50, 18, 32, 25] have shown significant improvement in action recognition.”
Asynchronous Temporal Fields for Action Recognition (2016, background): “Understanding activities and actions has an extensive history [32, 59, 22, 17, 23, 2, 26, 56, 29, 21].”
Detecting the Moment of Completion: Temporal Models for Localising Action Completion (2017, methodology): “For example, [6,7,8,9,10,11,12,13] extract space-time features from interest points or dense trajectories, encode them using BoWs or FVs, and then use SVM to recognise the actions.”
Robust Action Recognition Based on a Hierarchical Model (2013, methodology): “These methods can be divided into two categories: recognition based on global video descriptors [4] and recognition based on local video descriptors [1] [3].”
Action recognition in videos acquired by a moving camera using motion decomposition of Lagrangian particle trajectories (2011, methodology): “Action Ours Ours(All Trajs) STIP [30] Trajecton [17] Average 47.”
Modeling the Temporal Extent of Actions (2010, methodology): “As per [3] and [21], we use 4000 histogram bins for HOF and HOG space-time interest points, and 512 bins for Trajecton features.”
Depth Induced Feature Representation for 4D Human Activity Recognition (2014, methodology): “Roughly, we divide activity recognition techniques into four categories, Bag-ofFeatures/SVM (BoF/SVM) approaches [17], Deformable Part Models (DPM) approaches[39], silhouette representation [3], [5], [14], [22], feature trajectories [38], [21], [23], [31].”
What do 15,000 object categories tell us about classifying and localizing actions? (2015, unknown): “Such techniques include sampling at interest points [7, 23], densely [40, 54] or along dense trajectories [9, 17, 28, 50].”
Spatio-temporal Matching for Human Detection in Video (2014, background): “Compared to the sparser KLT-based trackers [27, 28], densely tracking the feature points guarantees a good coverage of foreground motion and improves the quality of the trajectories in the presence of fast irregular motions.”
Efﬁcient feature extraction, encoding and classiﬁcation for action recognition (unknown): “Several alternative local descriptors have been proposed including histograms of ﬂow orientations (HOF) [16], histograms of 3D gradients (HOG3D) [12, 33], motion boundary histograms (MBH) [4, 35], shapes of point trajectories [20, 26, 35], local trinary patterns [13, 37] and others.”
Discovering discriminative action parts from mid-level video representations (2012, methodology): “The first uses bag-of-words representations, where the “words” are computed either statically for each frame [25, 14] or from trajectories [27, 19, 22, 26].”
Better Exploiting Motion for Better Action Recognition (2013, background): “More recent techniques assume some underlying temporal motion model involving trajectories [2, 6, 7, 17, 18, 25, 29, 32].”
A Cause and Effect Analysis of Motion Trajectories for Modeling Actions (2014, methodology): “Similar to the proposed work, few of the trajectory-based methods to perform action classification are presented in [1, 17, 29, 25, 18, 11].”
A Robust and Efficient Video Representation for Action Recognition (2015, background): “Feature trajectories (Matikainen et al, 2009; Messing et al, 2009; Sun et al, 2009; Wang et al, 2013a) have been shown to be a good way for capturing the intrinsic dynamics of video data.”
Online robust action recognition based on a hierarchical model (2014, methodology): “Previous literature has mainly classified actions by the trajectories [1,2] or the distributions [3] of feature points.”
A temporal Bayesian model for classifying, detecting and localizing activities in video sequences (2012, methodology): “More recently descriptors based on tracking interest points have been explored [17, 16]; these use standard KLT trackers1 to track interest points.”