“…In [ 20 ], authors used 5361 labelled thermal face images captured using one type of a camera. Similarly in [ 36 ] 4200 images were used to creation a database, using only one type of a camera. In our work, 7920 thermal images of a face were collected with 10,555 face labels.…”
As the interest in facial detection grows, especially during a pandemic, solutions are sought that will be effective and bring more benefits. This is the case with the use of thermal imaging, which is resistant to environmental factors and makes it possible, for example, to determine the temperature based on the detected face, which brings new perspectives and opportunities to use such an approach for health control purposes. The goal of this work is to analyze the effectiveness of deep-learning-based face detection algorithms applied to thermal images, especially for faces covered by virus protective face masks. As part of this work, a set of thermal images was prepared containing over 7900 images of faces with and without masks. Selected raw data preprocessing methods were also investigated to analyze their influence on the face detection results. It was shown that the use of transfer learning based on features learned from visible light images results in mAP greater than 82% for half of the investigated models. The best model turned out to be the one based on Yolov3 model (mean average precision—mAP, was at least 99.3%, while the precision was at least 66.1%). Inference time of the models selected for evaluation on a small and cheap platform allows them to be used for many applications, especially in apps that promote public health.
“…In [ 20 ], authors used 5361 labelled thermal face images captured using one type of a camera. Similarly in [ 36 ] 4200 images were used to creation a database, using only one type of a camera. In our work, 7920 thermal images of a face were collected with 10,555 face labels.…”
As the interest in facial detection grows, especially during a pandemic, solutions are sought that will be effective and bring more benefits. This is the case with the use of thermal imaging, which is resistant to environmental factors and makes it possible, for example, to determine the temperature based on the detected face, which brings new perspectives and opportunities to use such an approach for health control purposes. The goal of this work is to analyze the effectiveness of deep-learning-based face detection algorithms applied to thermal images, especially for faces covered by virus protective face masks. As part of this work, a set of thermal images was prepared containing over 7900 images of faces with and without masks. Selected raw data preprocessing methods were also investigated to analyze their influence on the face detection results. It was shown that the use of transfer learning based on features learned from visible light images results in mAP greater than 82% for half of the investigated models. The best model turned out to be the one based on Yolov3 model (mean average precision—mAP, was at least 99.3%, while the precision was at least 66.1%). Inference time of the models selected for evaluation on a small and cheap platform allows them to be used for many applications, especially in apps that promote public health.
“…Most of the datasets have been collected in controlled environments. The VIS-TH dataset [13] contains 50 participants and 4,200 aligned thermal-visual image pairs with a resolution of 160 × 120. The dataset was acquired using a FLIR Duo R thermal camera in a controlled environment under various illumination conditions, facial expressions, head poses, and occlusions.…”
Face detection and localization of facial landmarks are the primary steps in building many face applications in computer vision. Numerous algorithms and benchmark datasets have been proposed to develop accurate face and facial landmark detection models in the visual domain. However, varying illumination conditions still pose challenging problems. Thermal cameras can address this problem because of their operation in longer wavelengths. However, thermal face detection and localization of facial landmarks in the wild condition are overlooked. The main reason is that most of the existing thermal face datasets have been collected in controlled environments. In addition, many of them contain no annotations of face bounding boxes and facial landmarks. In this work, we present a thermal face dataset with manually labeled bounding boxes and facial landmarks to address these problems. The dataset contains 9,202 images of 145 subjects, collected in both controlled and wild conditions. As a baseline, we trained the YOLOv5 object detection model and its adaptation for face detection, YOLO5Face, on our dataset. To show the efficacy of our dataset, we evaluated these models on the RWTH-Aachen thermal face dataset in addition to our test set. We have made the dataset, source code, and pretrained models publicly available at https://github.com/IS2AI/TFW to bolster research in thermal face analysis. <br>
“…Different from the Near-infrared collection devices which generate infrared light actively, thermal imagery systems detect the heat energy from the skin. The Thermal and Visible Paired Face Database (Mallat and Dugelay 2018) contains 2100 paired thermal-visible images from 50 subjects with different face and environment variations. Each subject participates in two acquisition sessions whose time interval is 3 to 4 months.…”
Near-infrared-visible (NIR-VIS) heterogeneous face recognition matches NIR to corresponding VIS face images. However, due to the sensing gap, NIR images often lose some identity information so that the NIR-VIS recognition issue is more difficult than conventional VIS face recognition. Recently, NIR-VIS heterogeneous face recognition has attracted considerable attention in the computer vision community because of its convenience and adaptability in practical applications. Various deep learning-based methods have been proposed and substantially increased the recognition performance, but the lack of NIR-VIS training samples leads to the difficulty of the model training process. In this paper, we propose a new Large-Scale Multi-Pose High-Quality NIR-VIS database 'LAMP-HQ' containing 56,788 NIR and 16,828 VIS images of 573 subjects with large diversities in pose, illumination, attribute, scene and accessory. We furnish a benchmark along with the protocol for NIR-VIS face recognition via generation on LAMP-HQ, including Pixel2-Pixel, CycleGAN, ADFL, PCFH, and PACH. Furthermore, we propose a novel exemplar-based variational spectral attention network to produce high-fidelity VIS images from NIR data. A spectral conditional attention module is introduced to reduce the domain gap between NIR and VIS data and then improve the performance of NIR-VIS heterogeneous face recognition on various databases including the LAMP-HQ.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.