Extended Co-occurrence HOG with Dense Trajectories for Fine-Grained Activity Recognition

Kataoka, Hirokatsu; Hashimoto, Kiyoshi; Iwata, Kenji; Satoh, Yutaka; Navab, Nassir; Ilić, Slobodan; Aoki, Yoshimitsu

doi:10.1007/978-3-319-16814-2_22

Cited by 14 publications

(19 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The baseline is the DT result, as reported in [17]. We also compare our results to those of [20] and [7] with the IXMAS and MPII datasets. Our method outperforms the state-of-the-art approaches.…”

Section: Comparison To the Baseline (Dtmentioning

confidence: 99%

“…However, thus far, the best approach for action recognition is arguably the DT approach , which is based on descriptions of the trajectories of tracked feature points, which are densely sampled. When obtaining these trajectories, the following spatiotemporal features are used: the trajectory histograms of oriented gradients ( Following the introduction of the original DT, dense sampling approaches for action recognition were also proposed in [15,5,14,7,18]. These studies improved the DT in various ways, for example, by introducing mid-level trajectory clustering ( Recent approaches have assigned human-object interactions to the IDT framework (Zhou et al, 2014(Zhou et al, , 2015.…”

Section: Space-time Action Featuresmentioning

confidence: 99%

“…Our general framework allows us to add extra features to the DCS. Here, we employ the co-occurrence feature suggested by [7]. This approach adds an extended cooccurrence HOG (ECoHOG) ( [8]) to the HOG, HOF, and MBH.…”

Section: Extra Feature Descriptor For Dcs and Ddtmentioning

confidence: 99%

See 2 more Smart Citations

Dominant Codewords Selection with Topic Model for Action Recognition

Kataoka

Iwata

Satoh

et al. 2016

2016 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self Cite

View full text Add to dashboard Cite

In this paper, we propose a framework for recognizing human activities that uses only in-topic dominant codewords and a mixture of intertopic vectors. Latent Dirichlet allocation (LDA) is used to develop approximations of human motion primitives; these are mid-level representations, and they adaptively integrate dominant vectors when classifying human activities. In LDA topic modeling, action videos (documents) are represented by a bag-of-words (input from a dictionary), and these are based on improved dense trajectories ([18]). The output topics correspond to human motion primitives, such as finger moving or subtle leg motion. We eliminate the impurities, such as missed tracking or changing light conditions, in each motion primitive. The assembled vector of motion primitives is an improved representation of the action. We demonstrate our method on four different datasets.

show abstract

Section: Comparison To the Baseline (Dtmentioning

confidence: 99%

Section: Space-time Action Featuresmentioning

confidence: 99%

See 1 more Smart Citation

Dominant Codewords Selection with Topic Model for Action Recognition

Kataoka

Iwata

Satoh

et al. 2016

2016 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Descriptors are applied to the densely captured trajectories by histograms of oriented gradients (HOG) [1], histograms of optical flow (HOF) [14], and motion boundary histograms (MBH) [2]. Dense sampling approaches for activity recognition were also proposed in [6,8,21] after the introduction of the first DT. These studies incremented DT, for example, by eliminating extra flow [6] and integrating a higher-order descriptor into the conventional features for fine-grained action recognition [8].…”

Section: Related Workmentioning

confidence: 99%

“…Dense sampling approaches for activity recognition were also proposed in [6,8,21] after the introduction of the first DT. These studies incremented DT, for example, by eliminating extra flow [6] and integrating a higher-order descriptor into the conventional features for fine-grained action recognition [8]. Additionally, Wang et al proposed an IDT [21] by executing camera motion estimation, canceling detection-based noise, and adding a Fisher vector [16].…”

Section: Related Workmentioning

confidence: 99%

Motion Representation with Acceleration Images

Kataoka

Shirakabe

et al. 2016

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Information of time differentiation is extremely important cue for a motion representation. We have applied first-order differential velocity from a positional information, moreover we believe that second-order differential acceleration is also a significant feature in a motion representation. However, an acceleration image based on a typical optical flow includes motion noises. We have not employed the acceleration image because the noises are too strong to catch an effective motion feature in an image sequence. On one hand, the recent convolutional neural networks (CNN) are robust against input noises.In this paper, we employ acceleration-stream in addition to the spatial-and temporal-stream based on the two-stream CNN. We clearly show the effectiveness of adding the acceleration stream to the two-stream CNN.

show abstract