Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses

U. M. Nunes, D. R. Faria, P. Peixoto

Research output: Contribution to conferencePaper

Abstract

Human activity recognition is still a very challenging
research area, due to the inherently complex temporal and
spatial patterns that characterize most human activities. This
paper proposes a human activity recognition framework based
on random forests, where each activity is classified requiring few
training examples (i.e. no frame-by-frame activity classification).
In a first approach, a simple mechanism that divides each action
sequence into a fixed-size window is employed, where max-min
skeleton-based features are extracted. In the second approach,
each window is delimited by a pair of automatically detected key
poses, where static and max-min dynamic features are extracted,
based on the determined activity example. Both approaches are
evaluated using the Cornell Activity Dataset [1], obtaining relevant
overall average results, considering that these approaches
are fast to train and require just a few training examples.
These characteristics suggest that the proposed framework can beuseful for real-time applications, where the activities are typicallywell distinctive and little training time is required, or to be
integrated in larger and sophisticated systems, for a first quick
impression/learning of certain activities
Original languageEnglish
Publication statusPublished - 31 Aug 2016
EventIEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR) - New York, United States
Duration: 26 Aug 201631 Aug 2016

Conference

ConferenceIEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR)
CountryUnited States
CityNew York
Period26/08/1631/08/16

Bibliographical note

© 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

Keywords

  • Human Daily Activity Recognition, Random Forest, Max-Min Skeleton-based Features, Key Poses, Static and Dynamic Features

Cite this

Nunes, U. M., Faria, D. R., & Peixoto, P. (2016). Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses. Paper presented at IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR), New York, United States.
Nunes, U. M. ; Faria, D. R. ; Peixoto, P. / Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses. Paper presented at IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR), New York, United States.
@conference{807557c99328440081cdaebd66d3e502,
title = "Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses",
abstract = "Human activity recognition is still a very challengingresearch area, due to the inherently complex temporal andspatial patterns that characterize most human activities. Thispaper proposes a human activity recognition framework basedon random forests, where each activity is classified requiring fewtraining examples (i.e. no frame-by-frame activity classification).In a first approach, a simple mechanism that divides each actionsequence into a fixed-size window is employed, where max-minskeleton-based features are extracted. In the second approach,each window is delimited by a pair of automatically detected keyposes, where static and max-min dynamic features are extracted,based on the determined activity example. Both approaches areevaluated using the Cornell Activity Dataset [1], obtaining relevantoverall average results, considering that these approachesare fast to train and require just a few training examples.These characteristics suggest that the proposed framework can beuseful for real-time applications, where the activities are typicallywell distinctive and little training time is required, or to beintegrated in larger and sophisticated systems, for a first quickimpression/learning of certain activities",
keywords = "Human Daily Activity Recognition, Random Forest, Max-Min Skeleton-based Features, Key Poses, Static and Dynamic Features",
author = "Nunes, {U. M.} and Faria, {D. R.} and P. Peixoto",
note = "{\circledC} 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. ; IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR) ; Conference date: 26-08-2016 Through 31-08-2016",
year = "2016",
month = "8",
day = "31",
language = "English",

}

Nunes, UM, Faria, DR & Peixoto, P 2016, 'Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses' Paper presented at IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR), New York, United States, 26/08/16 - 31/08/16, .

Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses. / Nunes, U. M.; Faria, D. R.; Peixoto, P.

2016. Paper presented at IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR), New York, United States.

Research output: Contribution to conferencePaper

TY - CONF

T1 - Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses

AU - Nunes, U. M.

AU - Faria, D. R.

AU - Peixoto, P.

N1 - © 2016 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

PY - 2016/8/31

Y1 - 2016/8/31

N2 - Human activity recognition is still a very challengingresearch area, due to the inherently complex temporal andspatial patterns that characterize most human activities. Thispaper proposes a human activity recognition framework basedon random forests, where each activity is classified requiring fewtraining examples (i.e. no frame-by-frame activity classification).In a first approach, a simple mechanism that divides each actionsequence into a fixed-size window is employed, where max-minskeleton-based features are extracted. In the second approach,each window is delimited by a pair of automatically detected keyposes, where static and max-min dynamic features are extracted,based on the determined activity example. Both approaches areevaluated using the Cornell Activity Dataset [1], obtaining relevantoverall average results, considering that these approachesare fast to train and require just a few training examples.These characteristics suggest that the proposed framework can beuseful for real-time applications, where the activities are typicallywell distinctive and little training time is required, or to beintegrated in larger and sophisticated systems, for a first quickimpression/learning of certain activities

AB - Human activity recognition is still a very challengingresearch area, due to the inherently complex temporal andspatial patterns that characterize most human activities. Thispaper proposes a human activity recognition framework basedon random forests, where each activity is classified requiring fewtraining examples (i.e. no frame-by-frame activity classification).In a first approach, a simple mechanism that divides each actionsequence into a fixed-size window is employed, where max-minskeleton-based features are extracted. In the second approach,each window is delimited by a pair of automatically detected keyposes, where static and max-min dynamic features are extracted,based on the determined activity example. Both approaches areevaluated using the Cornell Activity Dataset [1], obtaining relevantoverall average results, considering that these approachesare fast to train and require just a few training examples.These characteristics suggest that the proposed framework can beuseful for real-time applications, where the activities are typicallywell distinctive and little training time is required, or to beintegrated in larger and sophisticated systems, for a first quickimpression/learning of certain activities

KW - Human Daily Activity Recognition, Random Forest, Max-Min Skeleton-based Features, Key Poses, Static and Dynamic Features

UR - http://www.tc.columbia.edu/conferences/roman2016/

M3 - Paper

ER -

Nunes UM, Faria DR, Peixoto P. Human Activity Recognition using Max-Min Skeleton-based Features and Key Poses. 2016. Paper presented at IEEE RO-MAN'16: Workshop on Behavior Adaptation, Interaction and Learning for Assistive Robotics (BAILAR), New York, United States.