Event Detection in Crowded Videos

By Yan Ke, Rahul Sukthankar, Martial Hebert. IEEE International Conference on Computer Vision, 2007

441 citations recorded by the source corpus.

Real-world actions occur often in crowded, dynamic environments. This poses a difficult challenge for current approaches to video event detection because it is difficult to segment the actor from the background due to distracting motion from other objects in the scene. We propose a technique for event recognition in crowded videos that reliably identifies actions in the presence of partial occlusion and background clutter. Our approach is based on three key ideas: (1) we efficiently match the volumetric representation of an event against oversegmented spatio-temporal video volumes; (2) we augment our shape-based features using flow; (3) rather than treating an event template as an atomic entity, we separately match by parts (both in space and time), enabling robustness against occlusions and actor variability. Our experiments on human actions, such as picking up a dropped object or waving in a crowd show reliable detection with few false positives.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/53964e0ccc0412e2fbb2cdf3483e1f383208febe

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

Change-Point Detection in Time-Series Data by Direct Density-Ratio Estimation (2009, background): “This problem is referred to as change-point detection [34, 16, 22] or event detection [13], and covers a broad range of real-world problems such as fraud detection in cellular systems [24, 3], intrusion detection in computer networks [35], irregular-motion detection in vision systems [20], signal se”
SST: Single-Stream Temporal Action Proposals (2017, background): “A number of existing methods approach the problem of temporal action localization [19, 22, 26, 27, 31, 32, 35, 40, 24, 30, 38].”
Tube Convolutional Neural Network (T-CNN) for Action Detection in Videos (2017, background): “[15] present an approach for event detection in crowded videos.”
Recognizing actions by shape-motion prototype trees (2009, methodology): “Although some preliminary work has been done for recognizing actions in challengingmovie scenarios [10,12,13,17] or group actions in sports scenarios [14], robustly recognizing actions viewed against a dynamic varying background is still an important challenge.”
Predicting the Where and What of Actors and Actions through Online Action Localization (2016, background): “Offline Action Localization has received significant attention in the past few years [41, 38, 9, 5, 13, 35, 11].”
Learning Spatio-Temporal Structure from RGB-D Videos for Human Activity Detection and Anticipation (2013, background): “Some recent effort in recognizing actions from longer video sequences take an event detection approach (Ke et al., 2007; Simon et al., 2010; Nguyen et al., 2009), where they evaluate a classifier function at many different segments of the video and then predict the event presence in segments.”
Video-Based Human Behavior Understanding: A Survey (2013, methodology): “Instead of analyzing the behavior of whole bodies, an alternative is to perform parts-based analysis [125], [126], where template matching is combined with a pictorial structure model to detect and localize actions.”
Change-Point Detection in Time-Series Data by Relative Density-Ratio Estimation (2012, background): “This problem is referred to as change-point detection [34, 16, 22] or event detection [13], and covers a broad range of real-world problems such as fraud detection in cellular systems [24, 3], intrusion detection in computer networks [35], irregular-motion detection in vision systems [20], signal se”
Chaotic invariants of Lagrangian particle trajectories for anomaly detection in crowded scenes (2010, background): “Limiting our review of the literature to work directly related to analysis of crowded scenes, common features used for modeling crowded motions include optical flow [9, 10, 11, 12], gradient [13], spatiotemporal volume [14], etc.”
Domain-Adaptive Discriminative One-Shot Learning of Gestures (2014, background): “As a result, several recent works have attempted to learn gestures at the other extreme – from single training examples using one-shot learning [16,17,19,20,22,24,33].”
Trajectons: Action recognition through the motion analysis of tracked features (2009, methodology): “We propose two variants of trajectons, which differ in their construction of trajectory snippets.”
Learning latent temporal structure for complex event detection (2012, background): “Although some of the recent works have considered Internet videos, complex activity recognition tasks are typically already temporally localized [13, 16], and event detection tasks focus only on localizing well-defined primitive events [9].”
Discovering discriminative action parts from mid-level video representations (2012, methodology): “The second uses global spatiotemporal templates, such as motion history [3], spatiotemporal shapes [1], and other templates [13], that retain the spatial structure.”
Watch-n-patch: Unsupervised understanding of actions and relations (2015, background): “…them, the temporal structure of actions is often discovered by Markov models such as Hidden Markov Model (HMM) [34] and semi- Markov [12] [32], or by linear dynamical systems [3], or by hierarchical grammars [27] [37] [20] [39] [2], or by other spatiotemporal representations [15] [26] [17] [19].”
Parsing Videos of Actions with Segmental Grammars (2014, background): “Markov models: There exists a large body of literature on extensions to HMMs in which states generate variablelength sequences; these are sometimes called variablelength HMMs [4, 20], semi-Markov models [11, 22, 26], or segmental HMMs [6].”
Actionness Estimation Using Hybrid Fully Convolutional Networks (2016, methodology): “[18] utilized global template matching with the volume features for event detection.”
Sequential Max-Margin Event Detectors (2014, background): “[10] detected human actions in videos of crowds.”
Anticipating Human Activities Using Object Affordances for Reactive Robotic Response (2013, background): “Some recent effort in recognizing actions from longer video sequences take an event detection approach [15], [37], [38], [39], where they evaluate a classifier function at many different segments of the video and then predict event presence .”
Hidden Part Models for Human Action Recognition: Probabilistic versus Max Margin (2011, methodology): “Our work is partly inspired by a recent work in part-based event detection [26].”
Fast action proposals for human action detection and search (2015, background): “Other sliding-window based action detection approaches include [12, 22, 23].”