English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Human action recognition from well-segmented 3D skeleton data has beenintensively studied and has been attracting an increasing attention. Onlineaction detection goes one step further and is more challenging, whichidentifies the action type and localizes the action positions on the fly fromthe untrimmed stream data. In this paper, we study the problem of online actiondetection from streaming skeleton data. We propose a multi-task end-to-endJoint Classification-Regression Recurrent Neural Network to better explore theaction type and temporal localization information. By employing a jointclassification and regression optimization objective, this network is capableof automatically localizing the start and end points of actions moreaccurately. Specifically, by leveraging the merits of the deep Long Short-TermMemory (LSTM) subnetwork, the proposed model automatically captures the complexlong-range temporal dynamics, which naturally avoids the typical sliding windowdesign and thus ensures high computational efficiency. Furthermore, the subtaskof regression optimization provides the ability to forecast the action prior toits occurrence. To evaluate our proposed model, we build a large streamingvideo dataset with annotations. Experimental results on our dataset and thepublic G3D dataset both demonstrate very promising performance of our scheme.

Online Human Action Detection using Joint Classification-Regression Recurrent Neural Networks