Academic
Publications
Latent-Dynamic Discriminative Models for Continuous Gesture Recognition

Latent-Dynamic Discriminative Models for Continuous Gesture Recognition,10.1109/CVPR.2007.383299,Louis-philippe Morency,Ariadna Quattoni,Trevor Darrel

Latent-Dynamic Discriminative Models for Continuous Gesture Recognition   (Citations: 65)
BibTex | RIS | RefWorks Download
Many problems in vision involve the prediction of a class label for each frame in an unsegmented sequence. In this paper, we develop a discriminative framework for simulta- neous sequence segmentation and labeling which can cap- ture both intrinsic and extrinsic class dynamics. Our ap- proach incorporates hidden state variables which model the sub-structure of a class sequence and learn dynamics be- tween class labels. Each class label has a disjoint set of associated hidden states, which enables efficient training and inference in our model. We evaluated our method on the task of recognizing human gestures from unsegmented video streams and performed experiments on three different datasets of head and eye gestures. Our results demonstrate that our model compares favorably to Support Vector Ma- chines, Hidden Markov Models, and Conditional Random Fields on visual gesture recognition tasks.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.
    • ...Many other variants of HCRFs have been introduced since then [18], [12], [5], but most of these were tested only on single-signal pattern recognition tasks (e.g., POS tagging [8], object recognition [14], body gesture recognition [18], [12], and phone classification [5]) and paid less attention to dealing with noisy input signals...
    • ...Many other variants of HCRFs have been introduced since then [18], [12], [5], but most of these were tested only on single-signal pattern recognition tasks (e.g., POS tagging [8], object recognition [14], body gesture recognition [18], [12], and phone classification [5]) and paid less attention to dealing with noisy input signals...
    • ...In [12], Morency et al. presented an LDCRF that does not require its input sequence to be segmented, and showed that it is suitable for a number of gesture recognition tasks...

    Yale Songet al. Multi-signal gesture recognition using temporal smoothing hidden condi...

    • ...We compare the performance of our approach (eLB) with two state of the art methods: CRF [9] and LDCRF [5], and to the traditional level building (LB) approach from speech recognition research...
    • ...Although CRF [9] and LDCRF [5] have shown improved results for limited number of labels, in our experiments we had to use them for 40+ labels...

    Sudeep Sarkaret al. Segmentation-robust representations, matching, and modeling for sign l...

    • ...However, learning a DCRF model with hidden variables shall result in a very difficult optimize problem [29]...
    • ...Morency et al. [29] present a Latent-Dynamic Conditional Random Field (LDCRF) model, which also incorporates hiddendynamic variables for Continuous Gesture Recognition...
    • ...In fact, in order to make the training and inference phases of the LDCRF model tractable, Morency et al. [29] impose a constraint that each class label has a disjoint set of associated hidden state variables...
    • ...In contrast to Latent-Dynamic Conditional Random Fields (LDCRF)[29], SHDCRF proposes to learn the sparse relations between the hidden variables and intent labels instead of specifying by human in LDCRF...
    • ...Baselines. Our proposed SHDCRF model is compared with three baseline models, which are Support Vector Machine (SVM) [8], which assumes the queries in a user search session are independent, the classical Conditional Random Field (CRF) [20], which considers the sequential information and the Latent-Dynamic Conditional Random Fields (LDCRF) [29], which assigns a disjoint set of hidden state variables to each class label in advance...
    • ... LDCRF. The Latent-Dynamic Conditional Random Fields (LDCRF) [29] was trained by varying the number of hidden states per label, say, from 2 to 6 states per label, and the regularization term in LDCRF was determined by crossvalidation to achieve the best performance for comparative study...

    Yelong Shenet al. Sparse hidden-dynamics conditional random fields for user intent under...

    • ...Moreover, the latent dynamic CRF [35] recovers the label layer to a linear chain, which can be applied to predict labels over unsegmented sequences by relations between inter- and intraclass labels...

    Maodi Huet al. Gait-Based Gender Classification Using Mixed Conditional Random Field

    • ...Morency et al [14] model the dynamics between gesture labels by using a Latent-Dynamic CRF model...

    Prithviraj Banerjeeet al. Learning neighborhood cooccurrence statistics of sparse features for h...

Sort by: