Two-Frame Motion Estimation Based on Polynomial Expansion

By Gunnar Farnebäck. 2003

3,116 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, 3D Vision & Pose

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

In Media Technology

In 3D Vision & Pose

Citations in the corpus

SOCRATES: Introducing Depth in Visual Wildlife Monitoring Using Stereo Vision (2022, methodology): “…n)−D(x−mx, y −my, n− 1)| (1) where NT is equal to the number of frames in the input video, NP is the number of pixels in a single frame, D(x, y, n) is the scalar disparity at some pixel (x, y) at time n, and mx,my is the optical flow from frame n to frame n−1, calculated using (Farnebäck, 2003).”
Two-Stream Convolutional Networks for Action Recognition in Videos (2014, methodology): “ operates on multiple-frame dense optical ﬂow, which is typically computed in an energy minimisation framework by solving for a displacement ﬁeld (typically at multiple image scales). A common method [7], is to use an eight-parameter global motion model, with the energy obtained by approximating loc”
Video Paragraph Captioning Using Hierarchical Recurrent Neural Networks (2015, methodology): “We first use Optical Flow [13] to roughly detect a bounding box for the actor on each frame.”
Long-Term Temporal Convolutions for Action Recognition (2016, neutral): “To investigate the dependency of action recognition on the quality of motion estimation, we experiment with three types of ﬂow inputs obtained either directly from the video encoding, referred to as MPEG ﬂow [27], or from two optical ﬂow estimators, namely Farneback [28] and Brox [29].”
A Closer Look at Spatiotemporal Convolutions for Action Recognition (2017, methodology): “We use Farneback’s method [8] to compute optical flow because of its efficiency.”
P-CNN: Pose-Based CNN Features for Action Recognition (2015, methodology): “The method first densely samples points which are tracked using optical flow [15].”
Learning to Generate Long-term Future via Hierarchical Prediction (2017, methodology): “An optical ﬂow based baseline used the last observed optical ﬂow (Farneback, 2003) to move the pixels of the last observed frame into the future.”
Visual Attention Guided Deep Imitation Learning (2017, methodology): “, optical flow calculated using [3]) which is included since human gaze is sensitive to movement.”
Cascaded Boundary Regression for Temporal Action Detection (2017, methodology): “For the motion feature, we sample 6 consecutive frames at the center of a unit and calculate optical flows [4] between them; these flows are then fed into the pretrained BN-Inception model [9, 26] and the output of "global pool" layer is extracted.”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, background): “High Speed Flow: With some exceptions (Wulff & Black (2015); Timofte & Gool (2015); Weinzaepfel et al. (2013); Farneback (2003); Zach et al. (2007)) most of the optical flow approaches are very inefficient and can not be applied in real-time which is necessary for applications in autonomous driving.”
Trajectory-Based Modeling of Human Actions with Motion Reference Points (2012, methodology): “To compute the dense optical flow, the algorithm of [25] is adopted, which is available from the OpenCV library.”
SLAC: A Sparsely Labeled Dataset for Action Classification and Localization (2017, positive): “Second, I3D uses the TV-L1 optical flow [39], which is more expensive and accurate than the Farneback optical flow [5] adopted by our model.”
PatchBatch: A Batch Augmented Loss for Optical Flow (2015, methodology): “28.50% 116s DeepFlow [38] 29.18% 17s HS [35] 42.18% 2.6m DB-TV-L1 [40] 47.97% 16s HAOF [6] 50.29% 16.2s PolyExpand [14] 53.32% 1s for 32D. Based on these results, further improvements of our method’s accuracy are expected with larger patch and representation sizes.”
Multi-modal Egocentric Activity Recognition using Audio-Visual Features (2018, methodology): “Motion estimation is performed by Farneback Optical Flow algorithm [19].”
NoRMCorre: An online algorithm for piecewise rigid motion correction of calcium imaging data (2017, methodology): “The residual motion was calculated with the dense optical flow (OF) algorithm of Farnebäck [4] in its OpenCV (v3.”
Atari-HEAD: Atari Human Eye-Tracking and Demonstration Dataset (2019, methodology): “We also provide tools to extract optical ﬂow (Farneb¨ack 2003) and hand-crafted bottom-up saliency features (orientation and intensity) (Itti, Koch, and Niebur 1998).”