Ponències/Comunicacions de congressos
http://hdl.handle.net/2117/3950
2024-03-29T08:41:59ZTraining deep learning algorithms with multispectral dataset of skin lesions for the improvement of skin cancer diagnosis
http://hdl.handle.net/2117/404004
Training deep learning algorithms with multispectral dataset of skin lesions for the improvement of skin cancer diagnosis
Rey Barroso, Laura; Vilaseca Ricart, Meritxell; Royo Royo, Santiago; Puig Sardá, Susana; Malvehy Guilera, Josep; Pellacani, Giovanni; Lihacova, Ilze; Bondarenko, Andrey; Burgos Fernández, Francisco Javier
Dermatologists are starting to make use of Computer-Aided Diagnosis based on deep learning algorithms, which can provide them with an objective judgement during evaluation of equivocal lesions. DL algorithms can be trained to classify skin lesions with datasets of diverse nature like traditional RGB, clinical and dermoscopic images, or more experimentally, with images from other modalities, such as multispectral imaging. In this work, we have evaluated and customized the different DL approaches that exist in the state of the art to classify a dataset of +500 images acquired on skin lesions. The images were acquired with a staring multispectral imaging prototype in the visible and near-infrared ranges. The best results were obtained for a customized model VGG-16 that combined 3D convolutional layers, 3D maxpooling layers and dropout regularization, leading to an overall accuracy of 71%.
Copyright 2023 Society of Photo‑Optical Instrumentation Engineers (SPIE). One print or electronic copy may be made for personal use only. Systematic reproduction and distribution, duplication of any material in this publication for a fee or for commercial purposes, and modification of the contents of the publication are prohibited.
2024-03-08T13:20:33ZRey Barroso, LauraVilaseca Ricart, MeritxellRoyo Royo, SantiagoPuig Sardá, SusanaMalvehy Guilera, JosepPellacani, GiovanniLihacova, IlzeBondarenko, AndreyBurgos Fernández, Francisco JavierDermatologists are starting to make use of Computer-Aided Diagnosis based on deep learning algorithms, which can provide them with an objective judgement during evaluation of equivocal lesions. DL algorithms can be trained to classify skin lesions with datasets of diverse nature like traditional RGB, clinical and dermoscopic images, or more experimentally, with images from other modalities, such as multispectral imaging. In this work, we have evaluated and customized the different DL approaches that exist in the state of the art to classify a dataset of +500 images acquired on skin lesions. The images were acquired with a staring multispectral imaging prototype in the visible and near-infrared ranges. The best results were obtained for a customized model VGG-16 that combined 3D convolutional layers, 3D maxpooling layers and dropout regularization, leading to an overall accuracy of 71%.Multimodal imaging sensor based on lidar for advanced perception tasks
http://hdl.handle.net/2117/403578
Multimodal imaging sensor based on lidar for advanced perception tasks
Garcia Gómez, Pablo; Bernal Pérez, Eduard; Rodríguez Aramendía, Ana; Rodrigo Arcay, Noel; Riu Gras, Jordi; Royo Royo, Santiago
Automated systems increase their requirements in all fields, tightening their performance requirements in aspects like reliability, and ease of manipulation. Within this communication, we will present the development of a compact perception unit which includes a 3D lidar, RGB and thermal imaging for advanced perception purposes. The proposed unit intends to solve the usual hardware problems that software developers intend to solve in field applications. The basic features and performance of the system will be presented, and the applicability of the multimodal sensing approach presented to different applications in security, autonomous vehicles, and other application areas will be overviewed with examples
2024-03-01T09:34:18ZGarcia Gómez, PabloBernal Pérez, EduardRodríguez Aramendía, AnaRodrigo Arcay, NoelRiu Gras, JordiRoyo Royo, SantiagoAutomated systems increase their requirements in all fields, tightening their performance requirements in aspects like reliability, and ease of manipulation. Within this communication, we will present the development of a compact perception unit which includes a 3D lidar, RGB and thermal imaging for advanced perception purposes. The proposed unit intends to solve the usual hardware problems that software developers intend to solve in field applications. The basic features and performance of the system will be presented, and the applicability of the multimodal sensing approach presented to different applications in security, autonomous vehicles, and other application areas will be overviewed with examplesImage formation with lenses, pinholes and slits in an optical box
http://hdl.handle.net/2117/401469
Image formation with lenses, pinholes and slits in an optical box
Escofet Soteras, Jaume; Pérez Cabré, Elisabet
Images formed by systems with small apertures, such as pinholes and slits, have very weak illumination. It is advisable an object with a high luminance and with considerable size to see these images. Besides, the device that performs the image must be solid and stable. In this work, a wooden box has been designed to obtain these images and to see them comfortably. In order to illuminate the object intensely, an LED ceiling light has been used. Images with slits and pinholes have been formed in the box. Geometrical relationships between the object and image sizes have been established and theoretical and experimental resolutions for different pinholes have also been compared. The device has also been used to form images with lenses and compare them with those formed by pinholes.
2024-02-08T13:26:17ZEscofet Soteras, JaumePérez Cabré, ElisabetImages formed by systems with small apertures, such as pinholes and slits, have very weak illumination. It is advisable an object with a high luminance and with considerable size to see these images. Besides, the device that performs the image must be solid and stable. In this work, a wooden box has been designed to obtain these images and to see them comfortably. In order to illuminate the object intensely, an LED ceiling light has been used. Images with slits and pinholes have been formed in the box. Geometrical relationships between the object and image sizes have been established and theoretical and experimental resolutions for different pinholes have also been compared. The device has also been used to form images with lenses and compare them with those formed by pinholes.A LiDAR imaging system using temporal and polarization discrimination for turbid media
http://hdl.handle.net/2117/398682
A LiDAR imaging system using temporal and polarization discrimination for turbid media
Ballesta Garcia, Maria; Peña Gutiérrez, Sara; Rodríguez Aramendía, Ana; García Gómez, Pablo; Rodrigo Arcay, Noel; Royo Royo, Santiago
In recent times, there has been a surge of interest in LiDAR imaging systems, particularly in outdoor terrestrial applications associated with computer vision. However, a significant hurdle preventing their widespread implementation lies in their limited tolerance for adverse weather conditions, such as fog. To address this challenge, researchers have explored the capability of polarization in improving detection capabilities in such media. This paper explores the potential of LiDAR technology to obtain polarized images through fog and investigates the impact of fog on object detection using digitized temporal signals and point clouds. The study utilizes a LiDAR-polarized imaging system using circular polarization, which has been shown to enhance image contrast in highly-dispersive media. The analysis of the polarimetric information of the backscattered light signal in fog reveals its influence on object detection and evaluates the range difference between orthogonal polarimetric channels: coplanar and cross-configuration. The results demonstrate that cross-configuration detection provides larger range and more detailed point clouds compared to coplanar configuration, particularly benefiting metallic objects, for the same foggy conditions. By utilizing circularly polarized incident light and cross-configuration detection, the LiDAR system can improve the signal-to-noise ratio by filtering out the co-polarized fog responses. However, the range of the system may be reduced compared to nonpolarized detection. Overall, our findings indicate that utilizing a cross-polarization detection setup can effectively reduce the impact of fog backscatter while preserving the return signal from objects of interest in the majority of cases.
2023-12-21T15:09:55ZBallesta Garcia, MariaPeña Gutiérrez, SaraRodríguez Aramendía, AnaGarcía Gómez, PabloRodrigo Arcay, NoelRoyo Royo, SantiagoIn recent times, there has been a surge of interest in LiDAR imaging systems, particularly in outdoor terrestrial applications associated with computer vision. However, a significant hurdle preventing their widespread implementation lies in their limited tolerance for adverse weather conditions, such as fog. To address this challenge, researchers have explored the capability of polarization in improving detection capabilities in such media. This paper explores the potential of LiDAR technology to obtain polarized images through fog and investigates the impact of fog on object detection using digitized temporal signals and point clouds. The study utilizes a LiDAR-polarized imaging system using circular polarization, which has been shown to enhance image contrast in highly-dispersive media. The analysis of the polarimetric information of the backscattered light signal in fog reveals its influence on object detection and evaluates the range difference between orthogonal polarimetric channels: coplanar and cross-configuration. The results demonstrate that cross-configuration detection provides larger range and more detailed point clouds compared to coplanar configuration, particularly benefiting metallic objects, for the same foggy conditions. By utilizing circularly polarized incident light and cross-configuration detection, the LiDAR system can improve the signal-to-noise ratio by filtering out the co-polarized fog responses. However, the range of the system may be reduced compared to nonpolarized detection. Overall, our findings indicate that utilizing a cross-polarization detection setup can effectively reduce the impact of fog backscatter while preserving the return signal from objects of interest in the majority of cases.Prevención de accidentes en obra mediante visión 3D e inteligencia artificial
http://hdl.handle.net/2117/394645
Prevención de accidentes en obra mediante visión 3D e inteligencia artificial
Mas Giménez, Gerard de; Garcia Gómez, Pablo; Rubio Ponde, Marcela C.; Royo Royo, Santiago
La construcción es un sector laboral peligroso para los trabajadores debido a las situaciones de alto riesgo asociadas al trabajo con maquinaria pesada en ambientes hostiles. Para prevenir accidentes, se ha desarrollado un sistema de percepción multimodal que combina un LiDAR, una cámara térmica, y una cámara RGB. A través de la fusión de datos y el uso de YOLO, un algoritmo de Deep Learning para la detección de objetos, el sistema puede detectar situaciones de riesgo en obra a tiempo real. La red neuronal fue entrenada con más de 2.500 imágenes y 40.000 etiquetas producidas por el sistema multimodal. Los resultados de la validación demuestran la efectividad del sistema en la detección de situaciones de riesgo en tiempo real.
2023-10-04T17:15:51ZMas Giménez, Gerard deGarcia Gómez, PabloRubio Ponde, Marcela C.Royo Royo, SantiagoLa construcción es un sector laboral peligroso para los trabajadores debido a las situaciones de alto riesgo asociadas al trabajo con maquinaria pesada en ambientes hostiles. Para prevenir accidentes, se ha desarrollado un sistema de percepción multimodal que combina un LiDAR, una cámara térmica, y una cámara RGB. A través de la fusión de datos y el uso de YOLO, un algoritmo de Deep Learning para la detección de objetos, el sistema puede detectar situaciones de riesgo en obra a tiempo real. La red neuronal fue entrenada con más de 2.500 imágenes y 40.000 etiquetas producidas por el sistema multimodal. Los resultados de la validación demuestran la efectividad del sistema en la detección de situaciones de riesgo en tiempo real.Single image defogging and evaluation with deep neural networks without the need of ground truth
http://hdl.handle.net/2117/394640
Single image defogging and evaluation with deep neural networks without the need of ground truth
Mas Giménez, Gerard de; Garcia Gómez, Pablo; Casas Pla, Josep Ramon; Royo Royo, Santiago
Fog, haze, or smoke are usual atmospheric phenomena that dramatically compromise the overall visibility of any scene, critically affecting features such as illumination, contrast, and contour detection of objects. The decrease in visibility compromises the performance of computer vision algorithms such as pattern recognition and segmentation, some of them very relevant for decision-making in the field of autonomous vehicles. Several dehazing methods have been proposed that either need to estimate fog parameters through physical models or are statistically based. But physical parameters greatly depend on the scene conditions, and statistically based methods require large datasets of natural foggy images together with the original images without fog, i.e. the ground truth, for evaluation. Obtaining proper fog-less ground truth images for pixel-to-pixel evaluation is costly and time-consuming, and this fact hinders progress in the field. This paper aims to tackle this issue by proposing a gradient-based metrics for image defogging evaluation that does not need a ground truth image without fog or a physical model. A comparison of the proposed metrics with metrics already used in the NTIRE 2018 defogging challenge as well as several state-of-the-art defogging evaluation metrics is performed to prove its effectiveness in a general situation, showing comparable results to conventional metrics and an improvement in the no-reference scene. A Matlab implementation of the proposed metrics has been developed and it is open-sourced in a public GitHub repository.
2023-10-04T16:41:36ZMas Giménez, Gerard deGarcia Gómez, PabloCasas Pla, Josep RamonRoyo Royo, SantiagoFog, haze, or smoke are usual atmospheric phenomena that dramatically compromise the overall visibility of any scene, critically affecting features such as illumination, contrast, and contour detection of objects. The decrease in visibility compromises the performance of computer vision algorithms such as pattern recognition and segmentation, some of them very relevant for decision-making in the field of autonomous vehicles. Several dehazing methods have been proposed that either need to estimate fog parameters through physical models or are statistically based. But physical parameters greatly depend on the scene conditions, and statistically based methods require large datasets of natural foggy images together with the original images without fog, i.e. the ground truth, for evaluation. Obtaining proper fog-less ground truth images for pixel-to-pixel evaluation is costly and time-consuming, and this fact hinders progress in the field. This paper aims to tackle this issue by proposing a gradient-based metrics for image defogging evaluation that does not need a ground truth image without fog or a physical model. A comparison of the proposed metrics with metrics already used in the NTIRE 2018 defogging challenge as well as several state-of-the-art defogging evaluation metrics is performed to prove its effectiveness in a general situation, showing comparable results to conventional metrics and an improvement in the no-reference scene. A Matlab implementation of the proposed metrics has been developed and it is open-sourced in a public GitHub repository.Temporal behavior and processing of the LiDAR signal in fog
http://hdl.handle.net/2117/392267
Temporal behavior and processing of the LiDAR signal in fog
Ballesta Garcia, Maria; Rodríguez Aramendía, Ana; García Gómez, Pablo; Rodrigo Arcay, Noel; Royo Royo, Santiago
The interest in LiDAR imaging systems has recently increased in outdoor ground-based applications related to computer vision, in fields like autonomous vehicles. However, for the complete settling of the technology, there are still obstacles related to outdoor performance, being its use in adverse weather conditions one of the most challenging. When working in bad weather, data shown in point clouds is unreliable and its temporal behavior is unknown. We have designed, constructed, and tested a scanning-pulsed LiDAR imaging system with outstanding characteristics related to optoelectronic modifications, in particular including digitization capabilities of each of the pulses. The system performance was tested in a macro-scale fog chamber and, using the collected data, two relevant phenomena were identified: the backscattering signal of light that first interacts with the media and false-positive points that appear due to the scattering properties of the media. Digitization of the complete signal can be used to develop algorithms to identify and get rid of them. Our contribution is related to the digitization, analysis, and characterization of the acquired signal when steering to a target under foggy conditions, as well as the proposal of different strategies to improve point clouds generated in these conditions.
2023-07-26T10:31:30ZBallesta Garcia, MariaRodríguez Aramendía, AnaGarcía Gómez, PabloRodrigo Arcay, NoelRoyo Royo, SantiagoThe interest in LiDAR imaging systems has recently increased in outdoor ground-based applications related to computer vision, in fields like autonomous vehicles. However, for the complete settling of the technology, there are still obstacles related to outdoor performance, being its use in adverse weather conditions one of the most challenging. When working in bad weather, data shown in point clouds is unreliable and its temporal behavior is unknown. We have designed, constructed, and tested a scanning-pulsed LiDAR imaging system with outstanding characteristics related to optoelectronic modifications, in particular including digitization capabilities of each of the pulses. The system performance was tested in a macro-scale fog chamber and, using the collected data, two relevant phenomena were identified: the backscattering signal of light that first interacts with the media and false-positive points that appear due to the scattering properties of the media. Digitization of the complete signal can be used to develop algorithms to identify and get rid of them. Our contribution is related to the digitization, analysis, and characterization of the acquired signal when steering to a target under foggy conditions, as well as the proposal of different strategies to improve point clouds generated in these conditions.The Extremely Large Telescope (ELT) M1 Local Coherencer to phase mirror segments
http://hdl.handle.net/2117/386073
The Extremely Large Telescope (ELT) M1 Local Coherencer to phase mirror segments
Gaizka, Murga; Díaz Castillo, Alexander; Sanquirce García, Rubén; Royo Royo, Santiago; Pizarro Bondia, Carlos; González Jaio, Maialen; Lara Sáez, Elena; Vega Aguirrezabalaga, Borja; Rodrigo Arcay, Noel; Santos Vives, Pau-Antoni; Förster, Andreas; Schmid, Sebastian; Leveque, Samuel; Giton, Philippe; Filippi, Giorgio
The ELT M1 Local Coherencer is a non-contact metrology system aimed to simultaneously measure the relative pistons on the six sides of a target M1 segment with respect to neighboring ones (reference segments) with an accuracy below 300nm in a range of ±250µm. This measurement shall be performed while the Local Coherencer is supported by the M1 Segment Manipulator hanging from the M1 Segment Crane. IDOM has developed for the M1 Local Coherencer a lean, compact and robust solution featuring: - Six lightweight and compact Sensing Modules whose main system is a partially coherent light interferometer for the piston measurements that hugely simplifies image processing and avoids any ambiguity in the measurements. - Comprehensive and robust alignment detection and alignment compensation systems that ensure proper positioning and prevent apparent (bias) piston measurement errors. - A lean embodiment in which all the subsystems, including control and safety elements, are mounted on a single support structure and enclosed in the specified design volume, with no need to use the space reserved in the M1 Segment Manipulator - A solution largely based on small COTS and simple electronics, which account for ease of use, high reliability, easy replaceability and high durability of the system.
2023-04-11T07:14:17ZGaizka, MurgaDíaz Castillo, AlexanderSanquirce García, RubénRoyo Royo, SantiagoPizarro Bondia, CarlosGonzález Jaio, MaialenLara Sáez, ElenaVega Aguirrezabalaga, BorjaRodrigo Arcay, NoelSantos Vives, Pau-AntoniFörster, AndreasSchmid, SebastianLeveque, SamuelGiton, PhilippeFilippi, GiorgioThe ELT M1 Local Coherencer is a non-contact metrology system aimed to simultaneously measure the relative pistons on the six sides of a target M1 segment with respect to neighboring ones (reference segments) with an accuracy below 300nm in a range of ±250µm. This measurement shall be performed while the Local Coherencer is supported by the M1 Segment Manipulator hanging from the M1 Segment Crane. IDOM has developed for the M1 Local Coherencer a lean, compact and robust solution featuring: - Six lightweight and compact Sensing Modules whose main system is a partially coherent light interferometer for the piston measurements that hugely simplifies image processing and avoids any ambiguity in the measurements. - Comprehensive and robust alignment detection and alignment compensation systems that ensure proper positioning and prevent apparent (bias) piston measurement errors. - A lean embodiment in which all the subsystems, including control and safety elements, are mounted on a single support structure and enclosed in the specified design volume, with no need to use the space reserved in the M1 Segment Manipulator - A solution largely based on small COTS and simple electronics, which account for ease of use, high reliability, easy replaceability and high durability of the system.Frequency domain analysis and filter design of continuous wave frequency modulated optical feedback signal for photonic sensing
http://hdl.handle.net/2117/384889
Frequency domain analysis and filter design of continuous wave frequency modulated optical feedback signal for photonic sensing
Jha, Ajit; Cenkeramadi, Linga Reddy; Royo Royo, Santiago
Optical feedback (OF) refers to the beating of the frequency shifted back reflected laser emission from the remote (vibrating) object with that of incident radiation in the laser cavity cause rapid fluctuation of the laser emission. This fluctuations (also known as fringes or optical feedback signal (OFS)) contain the relevant information regarding the vibrating object. Optical feedback, and its associated variants e.g. continuous wave frequency modulated OF (CWFM-OF) has been extensively used for non-contact, non-destructive and self-aligned photonic sensing applications such as displacement, vibrations, ranging and imaging to name few. Despite its advantages, the optical feedback signal (OFS) require processing them in time and/or frequency domain to extract the relevant parameters. In this article, we present detailed frequency domain analysis of OFS. Among the many parameters we thoroughly investigate the effect of intensity modulation index (m) and external distance (L ext ) on the bandwidth of OFS and then determine the (active) filter parameters required to process them. Finally we discuss the trade-off between different parameters for optimal performance and signal processing of OFS.
2023-03-13T14:16:42ZJha, AjitCenkeramadi, Linga ReddyRoyo Royo, SantiagoOptical feedback (OF) refers to the beating of the frequency shifted back reflected laser emission from the remote (vibrating) object with that of incident radiation in the laser cavity cause rapid fluctuation of the laser emission. This fluctuations (also known as fringes or optical feedback signal (OFS)) contain the relevant information regarding the vibrating object. Optical feedback, and its associated variants e.g. continuous wave frequency modulated OF (CWFM-OF) has been extensively used for non-contact, non-destructive and self-aligned photonic sensing applications such as displacement, vibrations, ranging and imaging to name few. Despite its advantages, the optical feedback signal (OFS) require processing them in time and/or frequency domain to extract the relevant parameters. In this article, we present detailed frequency domain analysis of OFS. Among the many parameters we thoroughly investigate the effect of intensity modulation index (m) and external distance (L ext ) on the bandwidth of OFS and then determine the (active) filter parameters required to process them. Finally we discuss the trade-off between different parameters for optimal performance and signal processing of OFS.Game-based Learning in Computer Engineering: A Workshop
http://hdl.handle.net/2117/381238
Game-based Learning in Computer Engineering: A Workshop
Farreras Esclusa, Montserrat; Armengol Cebrian, Jesús; Bofill, Pau; Hernandez, Angels
Gamification has gained popularity in the last years, it is used in primary and secondary schools, as well as in companies
and universities (Call, 2021). Along with this growth in popularity the number of available computer tools that facilitate the
implementation of quizzes, competitions, simulations, WebQuests etc. has also grown.
Play engages students and enhances learning, however not all sorts of games are equally fruitful. As in (Dave Eng, 2019),
we make a distinction between gamification and game-based learning (GBL). An example of gamification is a contest
where students get points for solving the usual exercises of the subject matter. An example of game-based learning is an
escape room where students get involved in studying and solving subject matter problems to get the required hints to
continue the game. In this sense, game-based learning is an instance of problem-based learning (PBL) [Lima, 2017]. An
interesting reflection on GBL can be found in [Valero, 2018].
The main objective of GBL is to provide an active learning environment, where students need to learn and apply the subject
matter in order to participate and eventually solve the game. Compared with frontal lectures, active learning has been
shown to provide higher motivation and deeper learning [Call, 2021] [Lopez-Fernandez, 2021].
From a general point of view, we can say that learning takes place in five stages [Bofill, 2007]. Namely: motivation,
information retrieval, understanding, application (or practice) and reflection or feed-back. GBL, then, reinforces the
autonomous realization of each of these stages.
An escape room is a game where players must solve different puzzles and riddles in order to finish the game (in order to
escape from the room). Escape rooms have been used extensively in education, since they allow for the organization of
subject matter exercises in a pleasant way [Veldamp, 2020].
2023-01-26T15:20:10ZFarreras Esclusa, MontserratArmengol Cebrian, JesúsBofill, PauHernandez, AngelsGamification has gained popularity in the last years, it is used in primary and secondary schools, as well as in companies
and universities (Call, 2021). Along with this growth in popularity the number of available computer tools that facilitate the
implementation of quizzes, competitions, simulations, WebQuests etc. has also grown.
Play engages students and enhances learning, however not all sorts of games are equally fruitful. As in (Dave Eng, 2019),
we make a distinction between gamification and game-based learning (GBL). An example of gamification is a contest
where students get points for solving the usual exercises of the subject matter. An example of game-based learning is an
escape room where students get involved in studying and solving subject matter problems to get the required hints to
continue the game. In this sense, game-based learning is an instance of problem-based learning (PBL) [Lima, 2017]. An
interesting reflection on GBL can be found in [Valero, 2018].
The main objective of GBL is to provide an active learning environment, where students need to learn and apply the subject
matter in order to participate and eventually solve the game. Compared with frontal lectures, active learning has been
shown to provide higher motivation and deeper learning [Call, 2021] [Lopez-Fernandez, 2021].
From a general point of view, we can say that learning takes place in five stages [Bofill, 2007]. Namely: motivation,
information retrieval, understanding, application (or practice) and reflection or feed-back. GBL, then, reinforces the
autonomous realization of each of these stages.
An escape room is a game where players must solve different puzzles and riddles in order to finish the game (in order to
escape from the room). Escape rooms have been used extensively in education, since they allow for the organization of
subject matter exercises in a pleasant way [Veldamp, 2020].