Learning Spatio-Temporal Structure from RGB-D Videos for Human Activity Detection and Anticipation

By Hema Swetha Koppula, Ashutosh Saxena. 2013

223 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Model Architectures

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 Model Architectures

Citations in the corpus

A self-organizing neural network architecture for learning human-object interactions (2017, background): “). jects’ appearance as contextual information which is then concatenated with body motion features represented as a bag of words. On the other hand, best state-of-the-art results from Koppula et al. [56] reported 83.1% of accuracy, 87% of precision and 82.7% of recall. In their work, spatiotemporal”
RGB-D-based action recognition datasets: A survey (2016, neutral): “.6 & 91.7 3. 96.76 1. S 2. S 3. S 1. LOSubO(5-fold CV) 2. LOSubO(30-fold CV) &CS (half subjects training) 3. LOSubO CAD-120 [22] 1. QQSTR-gt-tracks [79] 2. Skeleton feature+HMMs [80] 3. ATCRF [81] 1. 95.2(Activity Acc.) 95.2(Activity Prec.)95(Activity Rec.) 2. 94.4(Activity Acc.) 91.6(Sub-ac”
Space-Time Representation of People Based on 3D Skeletal Data: A Review (2016, background): “[144] TVSRF Space Conc Manif Hand X X X X Koppula and Saxena [145] Temporal Relational Features Conc Lowlv Hand X Wu and Shao [146] EigenJoints Conc Lowlv Deep X X X X”
3D Human Activity Recognition with Reconfigurable Convolutional Neural Networks (2014, methodology): “ignment of actions by maximum margin temporal warping. Tang et al. [26] captured the latent temporal structures of 2D activities based on the variable-duration hidden Markov model. Koppula and Saxena [10] applied the Conditional Random Fields to model the sub-activities and aordances of the objects ”
Unsupervised Learning of Long-Term Motion Dynamics for Videos (2017, methodology): “Another stream of work uses probabilistic graphical models such as Hidden Markov Models (HMM) [49], Conditional Random Fields (CRF) [12] or Latent Dialect Allocation (LDA) [45] to capture spatial-temporal structures and learn the relations in activities from RGB-D videos.”
Structural-RNN: Deep Learning on Spatio-Temporal Graphs (2015, methodology): “Spatio-temporal graphs (st-graphs) are a popular [39, 37, 4, 11, 32, 65, 22] general tool for representing such high-level spatio-temporal structures.”
Anticipating Human Activities Using Object Affordances for Reactive Robotic Response (2013, background): “Improving temporal segmentation further improves the anticipation performance [19].”
Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis (2019, unknown): “Their implementations can be found in [61].”
A Deep Structured Model with Radius–Margin Bound for 3D Human Activity Recognition (2015, methodology): “Koppula and Saxena [18] applied the Conditional Random Fields to model the sub-activities and af-fordances of the objects for 3D activity recognition.”
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].”
Qualitative and Quantitative Spatio-temporal Relations in Daily Living Activity Recognition (2014, methodology): “Benchmark setting approaches developed by Koppula [10, 11], Rybok [32] have focused on modelling activities using generalized description of objects.”
On Human Motion Prediction Using Recurrent Neural Networks (2017, background): “wide variety of fields, such as action prediction for sociallyaware robotics [21], 3D people tracking within computer vision [13,43], motion generation for computer graphics [22] or modeling biological motion in psychology [42].”
Learning latent structure for activity recognition (2014, methodology): “The results show that our model performs better than the state-of-the-art approach [3], while the model is more efficient in inference.”
Dynamic Multiscale Graph Neural Networks for 3D Skeleton Based Human Motion Prediction (2020, unknown): “The related techniques can be widely applied to many computer vision and robotics scenarios, such as human-computer interaction [24, 23, 17, 13], autonomous driving [6], and pedestrian tracking [1, 15, 3].”
Ieee Transactions on Pattern Analysis and Machine Intelligence 1 3d Reasoning from Blocks to Stability (background): “…that integrating depth with color information improves the performances of many vision tasks, such as segmentation (in [7]), contour detection (in [24]), object recognition (in [25], [26], and [27]), scene labeling (in [28], [29], [30] and [31]), and activity detection (in [32], [33] and [34]).”
Skeleton-based Human Activity Recognition for Video Surveillance (2015, background): “In [20], Koppula and Saxena extended the work presented in [19] by detecting the past human activities as well as anticipating the future human activities using object affordances.”
“Important stuff, everywhere!” Activity recognition with salient proto-objects as context (2014, methodology): “*Note that [17] is using ground truth object labels and thus is not directly comparable to our approach.”