E-Locus - Institutional Repository of the University of Crete - PerLNet: Learning to localize multiple periodic activities in real-world videos

Home PerLNet: Learning to localize multiple periodic activities in real-world videos

Results - Details

[Add to Basket]

Identifier

000424543

Title

PerLNet: Learning to localize multiple periodic activities in real-world videos

Alternative Title

PerLNet: Εντοπισμός πολλαπλών περιοδικών δραστηριοτήτων σε πραγματικά βίντεο με χρήση τεχνικών μάθησης

Author

Καρβούνας, Γεώργιος Δ.

Thesis advisor

Αργυρός, Αντώνης

Reviewer

Τραχανιάς, Παναγιώτης
Ρούσσος, Αναστάσιος

Abstract

This thesis addresses the problem of temporal periodicity localization, i.e., the problem of identifying all segments of a video that contain some sort of periodic or repetitive motion. To do so, the proposed method represents a video by the matrix of pairwise frame distances. These distances are computed on frame representations obtained with a convolutional neural network. On top of this representation, we design, implement and evaluate PerLNet, a convolutional neural network that is able to classify a given frame as belonging (or not) to a periodic video segment. An important characteristic of the employed representation is that it permits the handling of videos and periodic segments of arbitrary number and duration. Furthermore, the proposed training process requires a relatively small number of annotated videos. The proposed method drops several of the limiting assumptions of existing approaches regarding the contents of the video and the types of the observed periodic actions. Experimental results on recent, publicly available datasets validate our design choices, verify the generalization potential of PerLNet and demonstrate its superior performance in comparison to the current state of the art.

Language

English

Subject

Computer vision

Deap learning

Periodic activities

Περιοδικές δραστηριότητες

Υπολογιστική όραση

Issue date

2019-03-29

Collection

School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses