Sorry, you need to enable JavaScript to visit this website.

facebooktwittermailshare

CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS

Abstract: 

Human action recognition has a wide range of applications including biometrics and surveillance. Existing methods mostly focus on a single modality, insufficient to characterize variations among different motions. To address this problem, we present a CNN-based human action recognition framework by fusing depth and skeleton modalities. The proposed Adaptive Multiscale Depth Motion Maps (AM-DMMs) are calculated from depth maps to capture shape, motion cues. Moreover, adaptive temporal windows ensure that AM-DMMs are robust to motion speed variations. A compact and effective method is also proposed to encode the spatio-temporal information of each skeleton sequence into three maps, referred to as Stable Joint Distance Maps (SJDMs) which describe different spatial relationships between the joints. A multi-channel CNN is adopted to exploit the discriminative features from texture color images encoded from AM-DMMs and SJDMs for effective recognition. The proposed method has been evaluated on UTD-MHAD Dataset and achieves the state-of-the-art result.

up
1 user has voted: Junyou He

Paper Details

Authors:
Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu
Submitted On:
20 November 2018 - 5:44am
Short Link:
Type:
Presentation Slides
Event:
Paper Code:
1466
Document Year:
2018
Cite

Document Files

CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS

(66)

Subscribe

[1] Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu, "CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS", IEEE SigPort, 2018. [Online]. Available: http://sigport.org/3692. Accessed: Jul. 20, 2019.
@article{3692-18,
url = {http://sigport.org/3692},
author = {Junyou He;Hailun Xia;Chunyan Feng;Yunfei Chu },
publisher = {IEEE SigPort},
title = {CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS},
year = {2018} }
TY - EJOUR
T1 - CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS
AU - Junyou He;Hailun Xia;Chunyan Feng;Yunfei Chu
PY - 2018
PB - IEEE SigPort
UR - http://sigport.org/3692
ER -
Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu. (2018). CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS. IEEE SigPort. http://sigport.org/3692
Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu, 2018. CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS. Available at: http://sigport.org/3692.
Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu. (2018). "CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS." Web.
1. Junyou He,Hailun Xia,Chunyan Feng,Yunfei Chu. CNN-BASED ACTION RECOGNITION USING ADAPTIVE MULTISCALE DEPTH MOTION MAPS AND STABLE JOINT DISTANCE MAPS [Internet]. IEEE SigPort; 2018. Available from : http://sigport.org/3692