Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition

Jian Liu, Hossein Rahmani, Naveed Akhtar, Ajmal Mian

Research output: Contribution to journalArticle

Abstract

We propose Human Pose Models that represent RGB and depth images of human poses independent of clothing textures, backgrounds, lighting conditions, body shapes and camera viewpoints. Learning such universal models requires training images where all factors are varied for every human pose. Capturing such data is prohibitively expensive. Therefore, we develop a framework for synthesizing the training data. First, we learn representative human poses from a large corpus of real motion captured human skeleton data. Next, we fit synthetic 3D humans with different body shapes to each pose and render each from 180 camera viewpoints while randomly varying the clothing textures, background and lighting. Generative Adversarial Networks are employed to minimize the gap between synthetic and real image distributions. CNN models are then learned that transfer human poses to a shared high-level invariant space. The learned CNN models are then used as invariant feature extractors from real RGB and depth frames of human action videos and the temporal variations are modelled by Fourier Temporal Pyramid. Finally, linear SVM is used for classification. Experiments on three benchmark human action datasets show that our algorithm outperforms existing methods by significant margins for RGB only and RGB-D action recognition.

Original languageEnglish
JournalInternational Journal of Computer Vision
DOIs
Publication statusPublished - 6 Aug 2019

Fingerprint

Textures
Lighting
Cameras
Experiments

Cite this

@article{1b71d6a5ffcb41629bf74b89bde20f72,
title = "Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition",
abstract = "We propose Human Pose Models that represent RGB and depth images of human poses independent of clothing textures, backgrounds, lighting conditions, body shapes and camera viewpoints. Learning such universal models requires training images where all factors are varied for every human pose. Capturing such data is prohibitively expensive. Therefore, we develop a framework for synthesizing the training data. First, we learn representative human poses from a large corpus of real motion captured human skeleton data. Next, we fit synthetic 3D humans with different body shapes to each pose and render each from 180 camera viewpoints while randomly varying the clothing textures, background and lighting. Generative Adversarial Networks are employed to minimize the gap between synthetic and real image distributions. CNN models are then learned that transfer human poses to a shared high-level invariant space. The learned CNN models are then used as invariant feature extractors from real RGB and depth frames of human action videos and the temporal variations are modelled by Fourier Temporal Pyramid. Finally, linear SVM is used for classification. Experiments on three benchmark human action datasets show that our algorithm outperforms existing methods by significant margins for RGB only and RGB-D action recognition.",
keywords = "CNN, Cross-subject, Cross-view, Depth sensor, GAN, Human action recognition",
author = "Jian Liu and Hossein Rahmani and Naveed Akhtar and Ajmal Mian",
year = "2019",
month = "8",
day = "6",
doi = "10.1007/s11263-019-01192-2",
language = "English",
journal = "International Journal of Computer Vision",
issn = "0920-5691",
publisher = "Springer",

}

TY - JOUR

T1 - Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition

AU - Liu, Jian

AU - Rahmani, Hossein

AU - Akhtar, Naveed

AU - Mian, Ajmal

PY - 2019/8/6

Y1 - 2019/8/6

N2 - We propose Human Pose Models that represent RGB and depth images of human poses independent of clothing textures, backgrounds, lighting conditions, body shapes and camera viewpoints. Learning such universal models requires training images where all factors are varied for every human pose. Capturing such data is prohibitively expensive. Therefore, we develop a framework for synthesizing the training data. First, we learn representative human poses from a large corpus of real motion captured human skeleton data. Next, we fit synthetic 3D humans with different body shapes to each pose and render each from 180 camera viewpoints while randomly varying the clothing textures, background and lighting. Generative Adversarial Networks are employed to minimize the gap between synthetic and real image distributions. CNN models are then learned that transfer human poses to a shared high-level invariant space. The learned CNN models are then used as invariant feature extractors from real RGB and depth frames of human action videos and the temporal variations are modelled by Fourier Temporal Pyramid. Finally, linear SVM is used for classification. Experiments on three benchmark human action datasets show that our algorithm outperforms existing methods by significant margins for RGB only and RGB-D action recognition.

AB - We propose Human Pose Models that represent RGB and depth images of human poses independent of clothing textures, backgrounds, lighting conditions, body shapes and camera viewpoints. Learning such universal models requires training images where all factors are varied for every human pose. Capturing such data is prohibitively expensive. Therefore, we develop a framework for synthesizing the training data. First, we learn representative human poses from a large corpus of real motion captured human skeleton data. Next, we fit synthetic 3D humans with different body shapes to each pose and render each from 180 camera viewpoints while randomly varying the clothing textures, background and lighting. Generative Adversarial Networks are employed to minimize the gap between synthetic and real image distributions. CNN models are then learned that transfer human poses to a shared high-level invariant space. The learned CNN models are then used as invariant feature extractors from real RGB and depth frames of human action videos and the temporal variations are modelled by Fourier Temporal Pyramid. Finally, linear SVM is used for classification. Experiments on three benchmark human action datasets show that our algorithm outperforms existing methods by significant margins for RGB only and RGB-D action recognition.

KW - CNN

KW - Cross-subject

KW - Cross-view

KW - Depth sensor

KW - GAN

KW - Human action recognition

UR - http://www.scopus.com/inward/record.url?scp=85070283916&partnerID=8YFLogxK

U2 - 10.1007/s11263-019-01192-2

DO - 10.1007/s11263-019-01192-2

M3 - Article

JO - International Journal of Computer Vision

JF - International Journal of Computer Vision

SN - 0920-5691

ER -