Fusing Spatiotemporal Features and Joints for 3D Action Recognition

By Y. Zhu, Wenbin Chen, Guo-qing Guo. 2013

238 citations recorded by the source corpus.

Subfields: Biomedical Engineering, Computer Vision and Pattern Recognition, Human-Computer Interaction

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

In Computer Vision and Pattern Recognition

In Human-Computer Interaction

Citations in the corpus

Differential Recurrent Neural Networks for Action Recognition (2015, background): “This resulted in a 136 dimensional feature vector per frame; (3) Offset: offset of the 3D joint positions between the current and the previous frame [38].”
When Kernel Methods Meet Feature Learning: Log-Covariance Network for Action Recognition From Skeletal Data (2017, neutral): “9% Spatio-temporal local features [44] 87.”
Skeleton-Based Action Recognition Using Spatio-Temporal LSTM Network with Trust Gates (2017, methodology): “Skeleton Joint Features [79] Geometric 87.”
Spatio-Temporal LSTM with Trust Gates for 3D Human Action Recognition (2016, methodology): “Experimental results on UT-Kinect Dataset (half-vs-half protocol [67]) Method Accuracy Skeleton Joint Features [67] 87.9% Lie Group [7] (reported by [68]) 93.6% Elastic functional coding [68] 94.9% ST-LSTM (Tree) + Trust Gate 95.0% [70] 96.1% Oﬂi et al. [71] 95.4% Vantigodi et al. [72] 97.6%…”
A novel hierarchical framework for human action recognition (2016, methodology): “It was used for evaluation in several 275 previous works [13, 16, 17, 19].”
Tensor Representations via Kernel Linearization for Action Recognition from 3D Skeletons (2016, neutral): “90% [35] 92 .”
Ensemble Deep Learning for Skeleton-Based Action Recognition Using Temporal Sliding LSTM Networks (2017, methodology): “UTKinect-Action dataset: We follow the protocol [25], in which half of the subjects are used for training and the remaining are used for testing.”
Transition Forests: Learning Discriminative Temporal Transitions for Action Recognition and Detection (2016, methodology): “Previous approaches encode the temporal variable in the feature space by stacking multiple frames [10], handcrafting temporal features [34, 40] or creating codebooks [34].”
Fusing shape and motion matrices for view invariant action recognition using 3D skeletons (2017, methodology): “The cross subjects setting follows [21]: subjects#1, 2, 3, 4, 5 for training and subjects #6, 7, 8, 9, 10 for testing.”
Spatiotemporal representation of 3D skeleton joints-based action recognition using modified spherical harmonics (2016, methodology): “The average accuracy of the proposed representation is 5.10% better than that given in (Zhu et al., 2013) and 2.08% better than that in (Xia et al., 2012).”
Fusing Geometric Features for Skeleton-Based Action Recognition Using Multilayer LSTM Networks (2018, unknown): “61% − − Skeleton Joint Features [44] − − − 87.”
Efficient Temporal Sequence Comparison and Classification Using Gram Matrix Embeddings on a Riemannian Manifold (2016, result): “Note that [42] and [36] also reported high accuracy with this dataset.”
Discriminative Orderlet Mining for Real-Time Recognition of Human-Object Interaction (2014, background): “Recently, with the development of the commodity depth sensors like Kinect, there has been a lot of interests in human action recognition from depth data [4, 12, 24].”
3D skeleton-based human action classification: A survey (2016, methodology): “Some methods such as [137,112,104] propose to use hybrid features where skeletal data and information extracted from depth maps and/or RGB video are jointly used for classifying the action sequence.”
WALKING WALKing walking: Action Recognition from Action Echoes (2017, methodology): “…with mean-pooling and softmax; 2) CNN alone with the fusion strategy MSSC; 3) CNN alone with MSMC. Figure 6 shows that muli-step-fusion based ConvESN Zhu et al. , 2013] 87.90 LTBSVM [Slama et al. , 2015] 88.50 HOJ3D+HMM [Xia et al. , 2012] 90.92 DP+KNN [Devanne et al. , 2013] 91.50 Skeletons Lie…”
Learning features combination for human action recognition from skeleton sequences (2017, methodology): “We consider the LOOCV scheme statistically more stable than the single cross-validation assessment employed by [32] and [8].”
Human Action Recognition From Various Data Modalities: A Review (2020, methodology): “In the work of [179], features were fused based on the random decision forests method.”