Evaluation of Local Spatio-temporal Features for Action Recognition

By Heng Wang, Muhammad Muneeb Ullah, Alexander Kläser, Ivan Laptev, Cordelia Schmid. 2009

1,506 citations recorded by the source corpus.

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

Disentangling Factors of Variation via Generative Entangling (2012, background): “This is the motivation behind both the inclusion of feature pooling stages in the convolutional network architecture (LeCun et al., 1989) and the recent trend toward representations based on large scale pooling of low-level features (Wang et al., 2009; Coates et al., 2011).”
One-shot-learning gesture recognition using HOG-HOF features (2013, background): “Wang et al. (2009) have evaluated a number of feature descriptors and bag-of-features models for action recognition.”
Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice (2014, neutral): “Local Tangent-based Coding [62] assumes that codebook and descriptors are embedded in a smooth manifold.”
Human Action Recognition Using Factorized Spatio-Temporal Convolutional Networks (2015, background): “Since video sequences can naturally be viewed as three-dimensional (3D) spatio-temporal signals, many existing methods seek to develop different spatiotemporal features for representing spatially and temporally coupled action patterns [12, 31, 8, 5].”
ActionVLAD: Learning Spatio-Temporal Aggregation for Action Classification (2017, background): “The descriptors are then aggregated into a bag-of-visualwords like representation resulting in a fixed-length descriptor vector for each video [52, 54].”
Human Action Recognition From Various Data Modalities: A Review (2020, background): “In [84], a few sparse methods and dense sampling methods were compared, where dense sampling was found to be able to outperform many other detectors.”
Representing Videos Using Mid-level Discriminative Patches (2013, background): “The second class of approaches is based on bag of features models [17, 24, 28], where sparse spatio-temporal interest points, dense interest points [32], page-rank features [21], or discriminative class-specific features [15], are computed as part of a bag of words representation on local features.”
Describing Videos by Exploiting Temporal Structure (2015, methodology): “Each cuboid is represented by concatenating the histograms of oriented gradients, oriented flow and motion boundary (HoG, HoF, MbH) [9, 40] with 33 bins.”
YouTube-8M: A Large-Scale Video Classification Benchmark (2016, methodology): “Inspired by the success of various classic bag of words representations for video classiﬁcation [23, 36], we next consider a Deep Bag-of-Frames (DBoF) approach.”
Compressed Video Action Recognition (2017, methodology): “Traditionally, for video action recognition, the community utilized hand-crafted features, such as Histogram of Oriented Gradients (HOG) [3] or Histogram of Optical Flow (HOF) [19], both sparsely [19] and densely [43] sampled.”
Non-intrusive human activity monitoring in a smart home environment (2013, methodology): “different kernels (such as χ(2) kernel [7], intersection kernel, and linear kernel [8], [9]) can be used for action clas sification.”
Action and Event Recognition with Fisher Vectors on a Compact Feature Set (2013, neutral): “Currently, most successful methods are based on some form of local space-time features; see [16, 43] for recent evaluation studies.”
Sparse Representation Based Fisher Discrimination Dictionary Learning for Image Classification (2014, background): “Recently sparse representation techniques have also led to promising results in face recognition (Wright et al. 2009b; Wang et al. 2009; Yang and Zhang 2010; Yang et al. 2011a,b), handwritten digit and texture classification (Huang and Aviyente 2006; Mairal et al. 2009; Ramirez et al. 2010; Yang et…”
Watching Unlabeled Video Helps Learn New Human Actions from Very Few Labeled Snapshots (2013, methodology): “This trick has even been employed to produce flipped versions of video sequences for activity recognition [31].”
Minimum Variance Extreme Learning Machine for human action recognition (2014, result): “We also compare the performance of the proposed MVELM network with that of the weighted kernel Support Vector Machine (kSVM) classifier [14, 15], which has been widely used in human action recognition.”