Tracking using Human Pose Matching with Deep Association Metric

This paper proposes a novel approach to track multiple people utilizing skeletal information combined with visual appearance features to improve the accuracy of tracking people across different frames of a video. We extracted the appearance feature vectors and skeletal feature vectors for each detec...

Full description

Saved in:
Bibliographic Details
Main Authors: Atishay Jain, Abhishek Dhiman, Balakrishna Pailla
Format: Article
Language:English
Published: LibraryPress@UF 2022-05-01
Series:Proceedings of the International Florida Artificial Intelligence Research Society Conference
Subjects:
Online Access:https://journals.flvc.org/FLAIRS/article/view/130580
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849763310287716352
author Atishay Jain
Abhishek Dhiman
Balakrishna Pailla
author_facet Atishay Jain
Abhishek Dhiman
Balakrishna Pailla
author_sort Atishay Jain
collection DOAJ
description This paper proposes a novel approach to track multiple people utilizing skeletal information combined with visual appearance features to improve the accuracy of tracking people across different frames of a video. We extracted the appearance feature vectors and skeletal feature vectors for each detected person in every frame. Each individual was tracked by considering the cosine distance between the skeletal feature vectors along with the euclidean distance between the appearance feature vectors across different frames of a video. This reduces the dependency of the tracker over appearances of people thus making it more consistent, especially in videos with people having similar appearances such as sports videos with players wearing similar jerseys. The stance of an individual in continuing frames is expected to be similar considering the high frame rate of modern camera devices. Therefore it is befitting to consider skeletal features along with appearance features for tracking. Our paper is an incremental paper demonstrating improvement over SORT with a deep association metric approach. Our approach utilizing skeletal information combined with visual appearance information returns better MOT results on the MOT17 dataset using the yolov3 detector.
format Article
id doaj-art-25fe941bbe4d489c952916e764553d23
institution DOAJ
issn 2334-0754
2334-0762
language English
publishDate 2022-05-01
publisher LibraryPress@UF
record_format Article
series Proceedings of the International Florida Artificial Intelligence Research Society Conference
spelling doaj-art-25fe941bbe4d489c952916e764553d232025-08-20T03:05:26ZengLibraryPress@UFProceedings of the International Florida Artificial Intelligence Research Society Conference2334-07542334-07622022-05-013510.32473/flairs.v35i.13058066779Tracking using Human Pose Matching with Deep Association MetricAtishay Jain0Abhishek DhimanBalakrishna PaillaVellore Institute of TechnologyThis paper proposes a novel approach to track multiple people utilizing skeletal information combined with visual appearance features to improve the accuracy of tracking people across different frames of a video. We extracted the appearance feature vectors and skeletal feature vectors for each detected person in every frame. Each individual was tracked by considering the cosine distance between the skeletal feature vectors along with the euclidean distance between the appearance feature vectors across different frames of a video. This reduces the dependency of the tracker over appearances of people thus making it more consistent, especially in videos with people having similar appearances such as sports videos with players wearing similar jerseys. The stance of an individual in continuing frames is expected to be similar considering the high frame rate of modern camera devices. Therefore it is befitting to consider skeletal features along with appearance features for tracking. Our paper is an incremental paper demonstrating improvement over SORT with a deep association metric approach. Our approach utilizing skeletal information combined with visual appearance information returns better MOT results on the MOT17 dataset using the yolov3 detector.https://journals.flvc.org/FLAIRS/article/view/130580computer visionmultiple object trackingdata associationhuman pose matchingneural networks
spellingShingle Atishay Jain
Abhishek Dhiman
Balakrishna Pailla
Tracking using Human Pose Matching with Deep Association Metric
Proceedings of the International Florida Artificial Intelligence Research Society Conference
computer vision
multiple object tracking
data association
human pose matching
neural networks
title Tracking using Human Pose Matching with Deep Association Metric
title_full Tracking using Human Pose Matching with Deep Association Metric
title_fullStr Tracking using Human Pose Matching with Deep Association Metric
title_full_unstemmed Tracking using Human Pose Matching with Deep Association Metric
title_short Tracking using Human Pose Matching with Deep Association Metric
title_sort tracking using human pose matching with deep association metric
topic computer vision
multiple object tracking
data association
human pose matching
neural networks
url https://journals.flvc.org/FLAIRS/article/view/130580
work_keys_str_mv AT atishayjain trackingusinghumanposematchingwithdeepassociationmetric
AT abhishekdhiman trackingusinghumanposematchingwithdeepassociationmetric
AT balakrishnapailla trackingusinghumanposematchingwithdeepassociationmetric