Exploring the Trade-off Between Accuracy and Observational Latency in Action Recognition

By Christopher Ellis, Syed Zain Masood, Marshall F. Tappen, J. Laviola, Rahul Sukthankar. 2012

256 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

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

In Artificial Intelligence

1. Wanqing Li
2. Pichao Wang
3. Jun Liu
4. Philip O. Ogunbona
5. Hao Zhang
6. Yingli Tian
7. Yonghong Hou
8. Fei Han

In Computer Vision and Pattern Recognition

1. Wanqing Li
2. Pichao Wang
3. Luc Van Gool
4. Jun Liu
5. G. Wang
6. Ming-Hsuan Yang
7. Dacheng Tao
8. Yingli Tian

Citations in the corpus

• RGB-D datasets using microsoft kinect or similar sensors: a survey (2016, background): “An application oriented case study about this dataset can be found in [23].”
• RGB-D-based action recognition datasets: A survey (2016, background): “UCFKinect In order to explore the trade-off between accuracy and observational latency when recognising actions, UCFKinect dataset [27] (http://www.”
• Space-Time Representation of People Based on 3D Skeletal Data: A Review (2016, methodology): “Many other datasets captured using a Kinect v1 camera were also released to the public, including the MSR Daily Activity 3D [121], MSR Action Pairs [113], Online RGBD Action (ORGBD) [107], UTKinect-Action [124], Florence 3D-Action [118], CMU-MAD [104], UTD-MHAD [103], G3D/G3Di [105], [119], SPHERE [”
• Dataset Augmentation in Feature Space (2017, methodology): “The final time series dataset we considered was the UCF Kinect action recognition dataset (Ellis et al., 2013).”
• Action Recognition Based on Joint Trajectory Maps Using Convolutional Neural Networks (2016, background): “be generally divided into three categories [54]: joints, groups of joints , and joint dynamics. Joint representation captures the correlation of the body joints by extracting spatial descriptor [30], [56]– poses [35], [61]–[63]. The groups of joints aim to detect the discriminative subsets of joints”
• Self-organizing neural integration of pose-motion features for human action recognition (2015, background): “When operating in complex environments, algorithms for the visual recognition of action classes face a trade-off between satisfactory accuracy and minimal recognition latency (Ellis et al., 2013).”
• Bio-inspired predictive orientation decomposition of skeleton trajectories for real-time human activity prediction (2015, methodology): “Logistic regression models [18] were employed to detect starting point of human activities.”
• Accurate 3D action recognition using learning on the Grassmann manifold (2015, methodology): “The same experimental setup as in Ellis et al. [41] is followed.”
• A discussion on the validation tests employed to compare human action recognition methods using the MSR Action3D dataset (2014, unknown): “[47] 2013 - 65.”
• Effective Active Skeleton Representation for Low Latency Human Action Recognition (2016, background): “Several features of the human skeleton have been proposed in the existing work, including Euclidean distance between each paired joints [3], [4], the histogram of 3D joints [5] and the path”
• Survey on Emotional Body Gesture Recognition (2018, unknown): “Sample upper and lower body postures [154].”
• Gesture Modeling by Hanklet-Based Hidden Markov Model (2014, background): “Methods [47]* [10]* [11]* [12] [24] [48]** [13]** [25] [23] [13]** Ours”
• Learning hierarchical spatio-temporal pattern for human activity prediction (2016, methodology): “We chose MSR-Action3D [18], 3D Online Action Dataset [30], UTKinect-Action [19], and UCF Kinect Dataset [31] for evaluating the early activity prediction accuracy and the full recognition accuracy.”
• Capturing Global and Local Dynamics for Human Action Recognition (2014, unknown): “[4].”
• Human Action Recognition by Shape Analysis of Motion Trajectories on Riemannian Manifold (2014, methodology): “, a 70/30 split protocol instead of the 4-fold cross validation proposed in [11]), 546 so their results are not reported here.”
• Online action recognition using covariance of shape and motion (2014, methodology): “In [23] the authors present a method for low latency online action recognition based on detecting distinctive canonical poses in the 3D motion of the subject’s skeleton.”
• Action Recognition Based on Joint Trajectory Maps with Convolutional Neural Networks (2016, background): “Joint representation captures the correlation of the body joints by extracting spatial descriptor [30], [56]– [59], geometric descriptor [33], [34], [39]–[41], [60] or key poses [35], [61]–[63].”
• Action Recognition Using Rate-Invariant Analysis of Skeletal Shape Trajectories (2016, methodology): “Some of these datasets are also used in the experimental studies described later.”