Behavior recognition via sparse spatio-temporal features

By Piotr Doll'ar, Vincent Rabaud, Gary W. Cottrell, Serge J. Belongie. 2005

2,847 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

Delving into Egocentric Actions (2015, neutral): “Using only egocentric features, we achieve a reasonable performance comparable to Cuboids.”
SlowFast Networks for Video Recognition (2018, neutral): “Actions can be formulated as spatiotemporal objects and captured by oriented filtering in spacetime, as done by HOG3D [31] and cuboids [10].”
Histogram of Oriented Principal Components for Cross-View Action Recognition (2014, methodology): “For example, Cheng et al. [3] used the Cuboid interest point detector [46] and proposed a Comparative Coding Descriptor (CCD).”
Action recognition with trajectory-pooled deep-convolutional descriptors (2015, methodology): “Space Time Interest Points [16] proposed Harris3D detector to extract informative regions, while Cuboid [7] detector relied on temporal Gabor filters.”
C3D: Generic Features for Video Analysis (2014, background): “Dollar et al. proposed Cuboids features for behavior recognition [11].”
A Closer Look at Spatiotemporal Convolutions for Action Recognition (2017, background): “Some proposed video representations include spatiotemporal interest points (STIPs) [21], SIFT-3D [27], HOG3D [18], Motion Boundary Histogram [5], Cuboids [6], and ActionBank [26].”
Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice (2014, neutral): “The main contents of LTC are manifold approximation and intrinsic dimensionality estimation.”
Unstructured human activity detection from RGBD images (2011, methodology): “We have also made the dataset and code available open-source at: http://pr.cs.cornell.edu/humanactivities”
Temporal Segment Networks for Action Recognition in Videos (2017, background): “In recent years, researchers have developed many different spatio-temporal feature detectors for video, such as 3D-Harris [35], 3D-Hessian [36], Cuboids [37], Dense Trajectories [38], Improved Trajecto-ries [2].”
A Biologically Inspired System for Action Recognition (2007, methodology): “While these approaches have been successful for the recognition of actions from articulated objects such as humans (see [9] for a review), they are not expected to be useful in the case of less articulated objects such as rodents [5].”
Better Exploiting Motion for Better Action Recognition (2013, methodology): “First, we briefly review the local descriptors [5, 13, 16, 29, 30] used here along with dense trajectories [29].”
A Robust and Efficient Video Representation for Action Recognition (2015, background): “Local space-time features (Dollár et al. 2005; Laptev 2005) have been shown to be advantageous in handling such datasets, as they allow to directly build efficient video representations without non-trivial pre-processing steps, such as object tracking or motion segmentation.”
Trajectory-Based Modeling of Human Actions with Motion Reference Points (2012, background): “The first category extracts/learns spatial-temporal local features where many efforts have been devoted to the design of good detectors/descriptors [1,7,8,9,10,11] or feature learning algorithms [12,13,14].”
A Hierarchical Model of Shape and Appearance for Human Action Classification (2007, methodology): “Motion features are obtained using the separable linear filter method in [8].”
Tensor Canonical Correlation Analysis for Action Classification (2007, methodology): “The competing methods are based on histogram representations of the local space-time interest points with SVM (Dollar et al [5], Schuldt et al.”
Discovering Motion Primitives for Unsupervised Grouping and One-Shot Learning of Human Actions, Gestures, and Expressions (2013, neutral): “On one end of this spectrum are interestpoint-based representations where a single descriptor encodes the appearance [27] or motion [22], [8] in very small x À y À t volumes, while, on the other hand, features based on actor silhouettes [7], contours [50], and space-time volumes [36] attempt to…”
MoSIFT: Recognizing Human Actions in Surveillance Videos (2009, methodology): “We first wanted to evaluate the utility of our MoSIFT interest point detector by comparing against the well-known temporal Gabor filter [4] and the 3D Harris corner detector [1].”