Learning spatiotemporal graphs of human activities

Brendel, William; Todorovic, Sinisa

doi:10.1109/iccv.2011.6126316

Cited by 202 publications

(159 citation statements)

References 22 publications

(33 reference statements)

Supporting

Mentioning

155

Contrasting

Unclassified

Order By: Relevance

“…Despite the fact that [6] can achieve fast speed, the proposed dynamic BoW based matching is not discriminative since it drops all the spatial information from the interest points. Similarly, in [12][5], the temporal models are not flexible to handle speed variations of the activity pattern. Third, a large training dataset will be needed to learn the activity model in [12] [6].…”

Section: Related Workmentioning

confidence: 99%

“…Similarly, in [12][5], the temporal models are not flexible to handle speed variations of the activity pattern. Third, a large training dataset will be needed to learn the activity model in [12] [6]. However, in some applications such as activity search, the amount of training data is extremely limited.…”

Section: Related Workmentioning

confidence: 99%

“…Activity classification on UT-Interaction dataset with 20% data for training We use the setting with 20% data for training and the other 80% for testing on UT-interaction dataset. This evaluation method has been used in [10] [12]. Since the training data is not sufficient, we build our random projection trees from the testing data.…”

Section: Rpt On the Testing Datamentioning

confidence: 99%

“…In [12], it learns a spatial-temporal graph model for each activity and classifies the testing video as the one with the smallest matching cost. In [5], temporal information is utilized to build the "string of feature graph".…”

Section: Related Workmentioning

confidence: 99%

“…There are several limitations in these matching based algorithms. First, the template matching algorithms, e.g., graph matching [12][5], are computationally intensive. For example, in order to use these template based methods for activity localization, we need to use sliding-windows to scan all the possible sub-volumes, which is an extremely large search space.…”

Section: Related Workmentioning

confidence: 99%

See 4 more Smart Citations

Propagative Hough Voting for Human Activity Recognition

Yuan

Liu

2012

Computer Vision – ECCV 2012

View full text Add to dashboard Cite

Abstract. Hough-transform based voting has been successfully applied to both object and activity detections. However, most current Hough voting methods will suffer when insufficient training data is provided. To address this problem, we propose propagative Hough voting for activity analysis. Instead of letting local features vote individually, we perform feature voting using random projection trees (RPT) which leverages the low-dimension manifold structure to match feature points in the highdimensional feature space. Our RPT can index the unlabeled testing data in an unsupervised way. After the trees are constructed, the label and spatial-temporal configuration information are propagated from the training samples to the testing data via RPT. The proposed activity recognition method does not rely on human detection and tracking, and can well handle the scale and intra-class variations of the activity patterns. The superior performances on two benchmarked activity datasets validate that our method outperforms the state-of-the-art techniques not only when there is sufficient training data such as in activity recognition, but also when there is limited training data such as in activity search with one query example.

show abstract