Dense Body Part Trajectories for Human Action Recognition

Ramana ORUGANTI, Ibrahim Hamed Ismail RADWAN, Roland GOECKE

Research output: A Conference proceeding or a Chapter in BookConference contribution

3 Citations (Scopus)
3 Downloads (Pure)

Abstract

Several techniques have been proposed for human action recognition from videos. It has been observed that incorporating mid-level viz. human body and/or high-level information viz. pose estimation in the computation of low-level features viz. trajectories yields the best performance in action recognition where full body is presumed. However, in datasets with a large number of classes, where the full body may not be visible at all times, incorporating such mid- and high-level information is unexplored. Moreover, changes and developments in any stage will require a recompute of all low-level features. We decouple mid-level and low-level feature computation and study on benchmark action recognition datasets such as UCF50, UCF101 and HMDB51, containing the largest number of action classes to date. Further, we employ a part-based model for human body part detection in frames statically, thus also investigating classes where the full body is not present. We also track dense regions around the detected human body parts by Hungarian particle linking,
thus minimising most of the wrongly detected body parts and enriching the mid-level information.
Original languageEnglish
Title of host publication2014 IEEE International Conference on Image Processing, ICIP 2014
Editors Pesquet-Popescu, Fowler
Place of PublicationParis, France
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages1465-1469
Number of pages5
ISBN (Electronic)9781479957514
ISBN (Print)9781479957514
DOIs
Publication statusPublished - 2014
Event2014 IEEE International Conference on Image Processing - Paris, Paris, France
Duration: 27 Oct 201430 Oct 2014

Publication series

Name2014 IEEE International Conference on Image Processing, ICIP 2014

Conference

Conference2014 IEEE International Conference on Image Processing
CountryFrance
CityParis
Period27/10/1430/10/14

Fingerprint

Trajectories

Cite this

ORUGANTI, R., RADWAN, I. H. I., & GOECKE, R. (2014). Dense Body Part Trajectories for Human Action Recognition. In Pesquet-Popescu, & Fowler (Eds.), 2014 IEEE International Conference on Image Processing, ICIP 2014 (pp. 1465-1469). [7025293] (2014 IEEE International Conference on Image Processing, ICIP 2014). Paris, France: IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICIP.2014.7025293
ORUGANTI, Ramana ; RADWAN, Ibrahim Hamed Ismail ; GOECKE, Roland. / Dense Body Part Trajectories for Human Action Recognition. 2014 IEEE International Conference on Image Processing, ICIP 2014. editor / Pesquet-Popescu ; Fowler. Paris, France : IEEE, Institute of Electrical and Electronics Engineers, 2014. pp. 1465-1469 (2014 IEEE International Conference on Image Processing, ICIP 2014).
@inproceedings{7e155dc1e8d64e26a7af3bdcb6eccfff,
title = "Dense Body Part Trajectories for Human Action Recognition",
abstract = "Several techniques have been proposed for human action recognition from videos. It has been observed that incorporating mid-level viz. human body and/or high-level information viz. pose estimation in the computation of low-level features viz. trajectories yields the best performance in action recognition where full body is presumed. However, in datasets with a large number of classes, where the full body may not be visible at all times, incorporating such mid- and high-level information is unexplored. Moreover, changes and developments in any stage will require a recompute of all low-level features. We decouple mid-level and low-level feature computation and study on benchmark action recognition datasets such as UCF50, UCF101 and HMDB51, containing the largest number of action classes to date. Further, we employ a part-based model for human body part detection in frames statically, thus also investigating classes where the full body is not present. We also track dense regions around the detected human body parts by Hungarian particle linking,thus minimising most of the wrongly detected body parts and enriching the mid-level information.",
keywords = "Hungarian particle linking, Large scale action recognition, dense trajectories, part-based model",
author = "Ramana ORUGANTI and RADWAN, {Ibrahim Hamed Ismail} and Roland GOECKE",
year = "2014",
doi = "10.1109/ICIP.2014.7025293",
language = "English",
isbn = "9781479957514",
series = "2014 IEEE International Conference on Image Processing, ICIP 2014",
publisher = "IEEE, Institute of Electrical and Electronics Engineers",
pages = "1465--1469",
editor = "Pesquet-Popescu and Fowler",
booktitle = "2014 IEEE International Conference on Image Processing, ICIP 2014",
address = "United States",

}

ORUGANTI, R, RADWAN, IHI & GOECKE, R 2014, Dense Body Part Trajectories for Human Action Recognition. in Pesquet-Popescu & Fowler (eds), 2014 IEEE International Conference on Image Processing, ICIP 2014., 7025293, 2014 IEEE International Conference on Image Processing, ICIP 2014, IEEE, Institute of Electrical and Electronics Engineers, Paris, France, pp. 1465-1469, 2014 IEEE International Conference on Image Processing, Paris, France, 27/10/14. https://doi.org/10.1109/ICIP.2014.7025293

Dense Body Part Trajectories for Human Action Recognition. / ORUGANTI, Ramana; RADWAN, Ibrahim Hamed Ismail; GOECKE, Roland.

2014 IEEE International Conference on Image Processing, ICIP 2014. ed. / Pesquet-Popescu; Fowler. Paris, France : IEEE, Institute of Electrical and Electronics Engineers, 2014. p. 1465-1469 7025293 (2014 IEEE International Conference on Image Processing, ICIP 2014).

Research output: A Conference proceeding or a Chapter in BookConference contribution

TY - GEN

T1 - Dense Body Part Trajectories for Human Action Recognition

AU - ORUGANTI, Ramana

AU - RADWAN, Ibrahim Hamed Ismail

AU - GOECKE, Roland

PY - 2014

Y1 - 2014

N2 - Several techniques have been proposed for human action recognition from videos. It has been observed that incorporating mid-level viz. human body and/or high-level information viz. pose estimation in the computation of low-level features viz. trajectories yields the best performance in action recognition where full body is presumed. However, in datasets with a large number of classes, where the full body may not be visible at all times, incorporating such mid- and high-level information is unexplored. Moreover, changes and developments in any stage will require a recompute of all low-level features. We decouple mid-level and low-level feature computation and study on benchmark action recognition datasets such as UCF50, UCF101 and HMDB51, containing the largest number of action classes to date. Further, we employ a part-based model for human body part detection in frames statically, thus also investigating classes where the full body is not present. We also track dense regions around the detected human body parts by Hungarian particle linking,thus minimising most of the wrongly detected body parts and enriching the mid-level information.

AB - Several techniques have been proposed for human action recognition from videos. It has been observed that incorporating mid-level viz. human body and/or high-level information viz. pose estimation in the computation of low-level features viz. trajectories yields the best performance in action recognition where full body is presumed. However, in datasets with a large number of classes, where the full body may not be visible at all times, incorporating such mid- and high-level information is unexplored. Moreover, changes and developments in any stage will require a recompute of all low-level features. We decouple mid-level and low-level feature computation and study on benchmark action recognition datasets such as UCF50, UCF101 and HMDB51, containing the largest number of action classes to date. Further, we employ a part-based model for human body part detection in frames statically, thus also investigating classes where the full body is not present. We also track dense regions around the detected human body parts by Hungarian particle linking,thus minimising most of the wrongly detected body parts and enriching the mid-level information.

KW - Hungarian particle linking

KW - Large scale action recognition

KW - dense trajectories

KW - part-based model

UR - http://www.scopus.com/inward/record.url?scp=84949927407&partnerID=8YFLogxK

U2 - 10.1109/ICIP.2014.7025293

DO - 10.1109/ICIP.2014.7025293

M3 - Conference contribution

SN - 9781479957514

T3 - 2014 IEEE International Conference on Image Processing, ICIP 2014

SP - 1465

EP - 1469

BT - 2014 IEEE International Conference on Image Processing, ICIP 2014

A2 - Pesquet-Popescu, null

A2 - Fowler, null

PB - IEEE, Institute of Electrical and Electronics Engineers

CY - Paris, France

ER -

ORUGANTI R, RADWAN IHI, GOECKE R. Dense Body Part Trajectories for Human Action Recognition. In Pesquet-Popescu, Fowler, editors, 2014 IEEE International Conference on Image Processing, ICIP 2014. Paris, France: IEEE, Institute of Electrical and Electronics Engineers. 2014. p. 1465-1469. 7025293. (2014 IEEE International Conference on Image Processing, ICIP 2014). https://doi.org/10.1109/ICIP.2014.7025293