Multi-Person Pose Estimation Using Thermal Images

Chen, I-Chien; Wang, Chang-Jen; Wen, Chao-Kai; Tzou, Shiow-Jyu

doi:10.1109/access.2020.3025413

Cited by 18 publications

(9 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Prior to the introduction of FIR-Human, previous works implementing pose estimation networks able to infer joint's position used RGB images for network training and small sets of labeled FIR images for performance testing. This is the case of the systems proposed in [53] and [54], where the network is evaluated using reduced FIR datasets specifically designed for that purpose.…”

Section: B Datasets Used For Human Pose Estimation System Trainingmentioning

confidence: 99%

Fall Detection in Low-Illumination Environments From Far-Infrared Images Using Pose Detection and Dynamic Descriptors

Gutiérrez,

Martin,

Rodriguez

et al. 2024

IEEE Access

View full text Add to dashboard Cite

In an increasingly aging world, the effort to automate tasks associated with the care of elderly dependent individuals becomes more and more relevant if quality care provision at sustainable costs is desired. One of the tasks susceptible to automation in this field is the automatic detection of falls. The research effort undertaken to develop automatic fall detection systems has been quite substantial and has resulted in reliable fall detection systems. However, individuals who could benefit from these systems only consider their use in certain scenarios. Among them, a relevant scenario is the one associated to semi-supervised patients during the night who wake up and get out of bed, usually disoriented, feeling an urgent need to go to the toilet. Under these circumstances, usually, the person is not supervised, and a fall could go unnoticed until the next morning, delaying the arrival of urgently needed assistance. In this scenario, associated with nighttime rest, the patient prioritizes comfort, and in this situation, body-worn sensors typical of wearable systems are not a good option. Environmental systems, particularly visual-based ones with cameras deployed in the patient's environment, could be the ideal option for this scenario. However, it is necessary to work with far-infrared (FIR) images in the low-light conditions of this environment. This work develops and implements, for the first time, a fall detection system that works with FIR imagery. The system integrates the output of a human pose estimation neural network with a detection methodology which uses the relative movement of the body's most important joints in order to determine whether a fall has taken place. The pose estimation neural networks used represent the most relevant architectures in this field and have been trained using the first large public labeled FIR dataset. Thus, we have developed the first vision-based fall detection system working on FIR imagery able to operate in conditions of absolute darkness whose performance indexes are equivalent to the ones of equivalent systems working on conventional RGB images.INDEX TERMS Computer vision, convolutional neural, fall detection, infrared imaging.

show abstract

Section: B Datasets Used For Human Pose Estimation System Trainingmentioning

confidence: 99%

Fall Detection in Low-Illumination Environments From Far-Infrared Images Using Pose Detection and Dynamic Descriptors

Gutiérrez,

Martin,

Rodriguez

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…The ThermalPose dataset [4] features 24,000 thermal images with a resolution of 80×20 pixels, coupled with their corresponding visual pairs. The dataset was collected in an indoor environment and contains images of both single and multiple people.…”

Section: Related Workmentioning

confidence: 99%

“…As a baseline, we trained and tested a series of YOLOv8-pose models on our dataset. Also, we tested the models on the UCH-Thermal-Pose [16] dataset to show the efficacy of our dataset. We have made the dataset, source code, and pre-trained models publicly available to stimulate research in this area.…”

Section: Introductionmentioning

confidence: 99%

OpenThermalPose: An Open-Source Annotated Thermal Human Pose Dataset and Initial YOLOv8-Pose Baselines

Kuzdeuov,

Taratynova,

Tleuliyev

et al. 2024

Preprint

View full text Add to dashboard Cite

Human pose estimation has a variety of applications in action recognition, human-robot interaction, motion capture, augmented reality, sports analytics, and healthcare. There is a substantial stream of datasets and deep learning-based models to attain robust human pose estimation within the visible domain. Nonetheless, there are certain obstacles in this domain, including insufficient illumination and privacy concerns. These issues can be addressed using thermal cameras. However, only a limited number of annotated thermal human pose datasets are available to train data-hungry deep learning models. In this regard, we introduce a novel open-source thermal human pose dataset named OpenThermalPose. The dataset contains 6,090 thermal images and 14,315 annotated human instances. The annotations include bounding boxes and 17 anatomical keypoints, following the annotation format of the MS COCO dataset. The dataset covers various fitness exercises, multiple-person activities, and outdoor walking in different locations and weather conditions. As a baseline, we trained and evaluated YOLOv8-pose models on our dataset. We have made the dataset, source code, and pre-trained models publicly available at https://github.com/IS2AI/OpenThermalPose to bolster research in this area.

show abstract

“…Methods for detecting human pose over thermal images hardly appear in the literature [25], [26], [2], [27]. One of the main reasons is the lack of databases of thermal images with annotated poses of humans.…”

Section: Human Pose Detection Using Thermal Imagesmentioning

confidence: 99%

“…The lack of large datasets containing thermal images of people is an obstacle to achieving high performance in detectors. A dataset containing paired visible and thermal images is introduced in [26], as well as a new network architecture for detecting people in the thermal domain. The dataset contains 24,000 pairs of thermal and visible images, captured in indoor environments.…”

Section: Human Pose Detection Using Thermal Imagesmentioning

confidence: 99%

Human Pose Estimation Using Thermal Images

2023

View full text Add to dashboard Cite

This study addresses the human pose estimation problem on thermal images using Convolutional Neural Networks and Vision Transformer architectures. To do this, eight human pose estimation methods designed for visible images were extended to be applied in the thermal domain. Due to the lack of large, representative datasets containing labeled thermal images, this extension requires transfer learning between the visible and the thermal domain, and a database for fine-tuning the networks in the thermal domain. Thus, it is proposed to train the networks using a grayscale version of the COCO dataset, and then fine-tune them in the thermal domain. Fine-tuning is carried out using the new UCH-Thermal-Pose database presented in this work. This database includes 600 thermal images for training, 200 for validation, and 104 for testing, all of them fully labeled. Moreover, in the paper, a comparative study of the eight extended deep-based methods for human pose detection is carried out. The UCH-Thermal-Pose database and the source code of all the methods compared are made available for future studies.

show abstract

Multi-Person Pose Estimation Using Thermal Images

Cited by 18 publications

References 18 publications

Fall Detection in Low-Illumination Environments From Far-Infrared Images Using Pose Detection and Dynamic Descriptors

Fall Detection in Low-Illumination Environments From Far-Infrared Images Using Pose Detection and Dynamic Descriptors

OpenThermalPose: An Open-Source Annotated Thermal Human Pose Dataset and Initial YOLOv8-Pose Baselines

Human Pose Estimation Using Thermal Images

Contact Info

Product

Resources

About