Robotic grasping in dynamic environments is still one of the main challenges in automation tasks. Advances in deep learning methods and computational power suggest that the problem of robotic grasping can be solved by using a huge amount of training data and deep networks. Despite these huge accomplishments, the acceptance and usage in real-world scenarios is still limited. This is mainly due to the fact that the collection of the training data is expensive, and that the trained network is a black box. While the collection of the training data can sometimes be facilitated by carrying it out in simulation, the trained networks, however, remain a black box. In this study, a three-step model is presented that profits both from the advantages of using a simulation approach and deep neural networks to identify and evaluate grasp points. In addition, it even offers an explanation for failed grasp attempts. The first step is to find all grasp points where the gripper can be lowered onto the table without colliding with the object. The second step is to determine, for the grasp points and gripper parameters from the first step, how the object moves while the gripper is closed. Finally, in the third step, for all grasp points from the second step, it is predicted whether the object slips out of the gripper during lifting. By this simplification, it is possible to understand for each grasp point why it is stable and – just as important – why others are unstable or not feasible. All of the models employed in each of the three steps and the resulting Overall Model are evaluated. The predicted grasp points from the Overall Model are compared to the grasp points determined analytically by a force-closure algorithm, to validate the stability of the predicted grasps.
{"title":"A three-step model for the detection of stable grasp points with machine learning","authors":"Constanze Schwan, W. Schenck","doi":"10.3233/ICA-210659","DOIUrl":"https://doi.org/10.3233/ICA-210659","url":null,"abstract":"Robotic grasping in dynamic environments is still one of the main challenges in automation tasks. Advances in deep learning methods and computational power suggest that the problem of robotic grasping can be solved by using a huge amount of training data and deep networks. Despite these huge accomplishments, the acceptance and usage in real-world scenarios is still limited. This is mainly due to the fact that the collection of the training data is expensive, and that the trained network is a black box. While the collection of the training data can sometimes be facilitated by carrying it out in simulation, the trained networks, however, remain a black box. In this study, a three-step model is presented that profits both from the advantages of using a simulation approach and deep neural networks to identify and evaluate grasp points. In addition, it even offers an explanation for failed grasp attempts. The first step is to find all grasp points where the gripper can be lowered onto the table without colliding with the object. The second step is to determine, for the grasp points and gripper parameters from the first step, how the object moves while the gripper is closed. Finally, in the third step, for all grasp points from the second step, it is predicted whether the object slips out of the gripper during lifting. By this simplification, it is possible to understand for each grasp point why it is stable and – just as important – why others are unstable or not feasible. All of the models employed in each of the three steps and the resulting Overall Model are evaluated. The predicted grasp points from the Overall Model are compared to the grasp points determined analytically by a force-closure algorithm, to validate the stability of the predicted grasps.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-07-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73527528","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
J. Hernández-Barragán, C. López-Franco, N. Arana-Daniel, A. Alanis, Adriana Lopez-Franco
The inverse kinematics of robotic manipulators consists of finding a joint configuration to reach a desired end-effector pose. Since inverse kinematics is a complex non-linear problem with redundant solutions, sophisticated optimization techniques are often required to solve this problem; a possible solution can be found in metaheuristic algorithms. In this work, a modified version of the firefly algorithm for multimodal optimization is proposed to solve the inverse kinematics. This modified version can provide multiple joint configurations leading to the same end-effector pose, improving the classic firefly algorithm performance. Moreover, the proposed approach avoids singularities because it does not require any Jacobian matrix inversion, which is the main problem of conventional approaches. The proposed approach can be implemented in robotic manipulators composed of revolute or prismatic joints of n degrees of freedom considering joint limits constrains. Simulations with different robotic manipulators show the accuracy and robustness of the proposed approach. Additionally, non-parametric statistical tests are included to show that the proposed method has a statistically significant improvement over other multimodal optimization algorithms. Finally, real-time experiments on five degrees of freedom robotic manipulator illustrate the applicability of this approach.
{"title":"A modified firefly algorithm for the inverse kinematics solutions of robotic manipulators","authors":"J. Hernández-Barragán, C. López-Franco, N. Arana-Daniel, A. Alanis, Adriana Lopez-Franco","doi":"10.3233/ICA-210660","DOIUrl":"https://doi.org/10.3233/ICA-210660","url":null,"abstract":"The inverse kinematics of robotic manipulators consists of finding a joint configuration to reach a desired end-effector pose. Since inverse kinematics is a complex non-linear problem with redundant solutions, sophisticated optimization techniques are often required to solve this problem; a possible solution can be found in metaheuristic algorithms. In this work, a modified version of the firefly algorithm for multimodal optimization is proposed to solve the inverse kinematics. This modified version can provide multiple joint configurations leading to the same end-effector pose, improving the classic firefly algorithm performance. Moreover, the proposed approach avoids singularities because it does not require any Jacobian matrix inversion, which is the main problem of conventional approaches. The proposed approach can be implemented in robotic manipulators composed of revolute or prismatic joints of n degrees of freedom considering joint limits constrains. Simulations with different robotic manipulators show the accuracy and robustness of the proposed approach. Additionally, non-parametric statistical tests are included to show that the proposed method has a statistically significant improvement over other multimodal optimization algorithms. Finally, real-time experiments on five degrees of freedom robotic manipulator illustrate the applicability of this approach.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-06-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76608378","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
It is a fact that natural disasters often cause severe damage both to ecosystems and humans. Moreover, man-made disasters can have enormous moral and economic consequences for people. A typical example is the large deadly and catastrophic explosion in Beirut on 4 August 2020, which destroyed a very large area of the city. This research paper introduces a Geo-AI disaster response computer vision system, capable to map an area using material from Synthetic Aperture Radar (SAR). SAR is a unique form of radar that can penetrate the clouds and collect data day and night under any weather conditions. Specifically, the Memory-Augmented Deep Convolutional Echo State Network (MA/DCESN) is introduced for the first time in the literature, as an advanced Machine Vision (MAV) architecture. It uses a meta-learning technique, which is based on a memory-augmented approach. The target is the employment of Deep Reservoir Computing (DRC) for domain adaptation. The developed Deep Convolutional Echo State Network (DCESN) combines a classic Convolutional Neural Network (CNN), with a Deep Echo State Network (DESN), and analog neurons with sparse random connections. Its training is performed following the Recursive Least Square (RLS) method. In addition, the integration of external memory allows the storage of useful data from past processes, while facilitating the rapid integration of new information, without the need for retraining. The proposed DCESN implements a set of original modifications regarding training setting, memory retrieval mechanisms, addressing techniques, and ways of assigning attention weights to memory vectors. As it is experimentally shown, the whole approach produces remarkable stability, high generalization efficiency and significant classification accuracy, significantly extending the state-of-the-art Machine Vision methods.
{"title":"Geo-AI to aid disaster response by memory-augmented deep reservoir computing","authors":"Konstantinos Demertzis, L. Iliadis, E. Pimenidis","doi":"10.3233/ICA-210657","DOIUrl":"https://doi.org/10.3233/ICA-210657","url":null,"abstract":"It is a fact that natural disasters often cause severe damage both to ecosystems and humans. Moreover, man-made disasters can have enormous moral and economic consequences for people. A typical example is the large deadly and catastrophic explosion in Beirut on 4 August 2020, which destroyed a very large area of the city. This research paper introduces a Geo-AI disaster response computer vision system, capable to map an area using material from Synthetic Aperture Radar (SAR). SAR is a unique form of radar that can penetrate the clouds and collect data day and night under any weather conditions. Specifically, the Memory-Augmented Deep Convolutional Echo State Network (MA/DCESN) is introduced for the first time in the literature, as an advanced Machine Vision (MAV) architecture. It uses a meta-learning technique, which is based on a memory-augmented approach. The target is the employment of Deep Reservoir Computing (DRC) for domain adaptation. The developed Deep Convolutional Echo State Network (DCESN) combines a classic Convolutional Neural Network (CNN), with a Deep Echo State Network (DESN), and analog neurons with sparse random connections. Its training is performed following the Recursive Least Square (RLS) method. In addition, the integration of external memory allows the storage of useful data from past processes, while facilitating the rapid integration of new information, without the need for retraining. The proposed DCESN implements a set of original modifications regarding training setting, memory retrieval mechanisms, addressing techniques, and ways of assigning attention weights to memory vectors. As it is experimentally shown, the whole approach produces remarkable stability, high generalization efficiency and significant classification accuracy, significantly extending the state-of-the-art Machine Vision methods.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89578878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Serafín Alonso Castro, Antonio Morán Álvarez, Daniel Pérez, M. A. Prada, J. J. Fuertes, M. Domínguez
Global natural resources are affected by several causes such as climate change effects or unsustainable management strategies. Indeed, the use of water has been intensified in urban buildings because of the proliferation of HVAC (Heating, Ventilating and Air Conditioning) systems, for instance cooling towers, where an abundant amount of water is lost during the evaporation process. The measurement of the evaporation is challenging, so a virtual sensor could be used to tackle it, allowing to monitor and manage the water consumption in different scenarios and helping to plan efficient operation strategies which reduce the use of fresh water. In this paper, a deep generative approach is proposed for developing a virtual sensor for probabilistic estimation of the evaporation in cooling towers, given the surrounding conditions. It is based on a conditioned generative adversarial network (cGAN), whose generator includes a recurrent layer (GRU) that models the temporal information by learning from previous states and a densely connected layer that models the fluctuations of the conditions. The proposed deep generative approach is not only able to yield the estimated evaporation value but it also produces a whole probability distribution, considering any operating scenario, so it is possible to know the confidence interval in which the estimation is likely found. This deep generative approach is assessed and compared with other probabilistic state-of-the-art methods according to several metrics (CRPS, MAPE and RMSE) and using real data from a cooling tower located at a hospital building. The results obtained show that, to the best of our knowledge, our proposal is a noteworthy method to develop a virtual sensor, taking as input the current and last samples, since it provides an accurate estimation of the evaporation with wide enough confidence intervals, contemplating potential fluctuations of the conditions.
{"title":"Virtual sensor for probabilistic estimation of the evaporation in cooling towers","authors":"Serafín Alonso Castro, Antonio Morán Álvarez, Daniel Pérez, M. A. Prada, J. J. Fuertes, M. Domínguez","doi":"10.3233/ICA-210654","DOIUrl":"https://doi.org/10.3233/ICA-210654","url":null,"abstract":"Global natural resources are affected by several causes such as climate change effects or unsustainable management strategies. Indeed, the use of water has been intensified in urban buildings because of the proliferation of HVAC (Heating, Ventilating and Air Conditioning) systems, for instance cooling towers, where an abundant amount of water is lost during the evaporation process. The measurement of the evaporation is challenging, so a virtual sensor could be used to tackle it, allowing to monitor and manage the water consumption in different scenarios and helping to plan efficient operation strategies which reduce the use of fresh water. In this paper, a deep generative approach is proposed for developing a virtual sensor for probabilistic estimation of the evaporation in cooling towers, given the surrounding conditions. It is based on a conditioned generative adversarial network (cGAN), whose generator includes a recurrent layer (GRU) that models the temporal information by learning from previous states and a densely connected layer that models the fluctuations of the conditions. The proposed deep generative approach is not only able to yield the estimated evaporation value but it also produces a whole probability distribution, considering any operating scenario, so it is possible to know the confidence interval in which the estimation is likely found. This deep generative approach is assessed and compared with other probabilistic state-of-the-art methods according to several metrics (CRPS, MAPE and RMSE) and using real data from a cooling tower located at a hospital building. The results obtained show that, to the best of our knowledge, our proposal is a noteworthy method to develop a virtual sensor, taking as input the current and last samples, since it provides an accurate estimation of the evaporation with wide enough confidence intervals, contemplating potential fluctuations of the conditions.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-04-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78777486","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
D. Avola, Marco Cascio, L. Cinque, G. Foresti, D. Pannone
In recent years, the spread of video sensor networks both in public and private areas has grown considerably. Smart algorithms for video semantic content understanding are increasingly developed to support human operators in monitoring different activities, by recognizing events that occur in the observed scene. With the term event, we refer to one or more actions performed by one or more subjects (e.g., people or vehicles) acting within the same observed area. When these actions are performed by subjects that do not interact with each other, the events are usually classified as simple. Instead, when any kind of interaction occurs among subjects, the involved events are typically classified as complex. This survey starts by providing the formal definitions of both scene and event, and the logical architecture for a generic event recognition system. Subsequently, it presents two taxonomies based on features and machine learning algorithms, respectively, which are used to describe the different approaches for the recognition of events within a video sequence. This paper also discusses key works of the current state-of-the-art of event recognition, providing the list of datasets used to evaluate the performance of reported methods for video content understanding.
{"title":"Machine learning for video event recognition","authors":"D. Avola, Marco Cascio, L. Cinque, G. Foresti, D. Pannone","doi":"10.3233/ICA-210652","DOIUrl":"https://doi.org/10.3233/ICA-210652","url":null,"abstract":"In recent years, the spread of video sensor networks both in public and private areas has grown considerably. Smart algorithms for video semantic content understanding are increasingly developed to support human operators in monitoring different activities, by recognizing events that occur in the observed scene. With the term event, we refer to one or more actions performed by one or more subjects (e.g., people or vehicles) acting within the same observed area. When these actions are performed by subjects that do not interact with each other, the events are usually classified as simple. Instead, when any kind of interaction occurs among subjects, the involved events are typically classified as complex. This survey starts by providing the formal definitions of both scene and event, and the logical architecture for a generic event recognition system. Subsequently, it presents two taxonomies based on features and machine learning algorithms, respectively, which are used to describe the different approaches for the recognition of events within a video sequence. This paper also discusses key works of the current state-of-the-art of event recognition, providing the list of datasets used to evaluate the performance of reported methods for video content understanding.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-04-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76356964","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Daniel Saranovic, M. Pavlovski, W. Power, Ivan Stojkovic, Z. Obradovic
As the prevalence of drones increases, understanding and preparing for possible adversarial uses of drones and drone swarms is of paramount importance. Correspondingly, developing defensive mechanisms in which swarms can be used to protect against adversarial Unmanned Aerial Vehicles (UAVs) is a problem that requires further attention. Prior work on intercepting UAVs relies mostly on utilizing additional sensors or uses the Hamilton-Jacobi-Bellman equation, for which strong conditions need to be met to guarantee the existence of a saddle-point solution. To that end, this work proposes a novel interception method that utilizes the swarm’s onboard PID controllers for setting the drones’ states during interception. The drone’s states are constrained only by their physical limitations, and only partial feedback of the adversarial drone’s positions is assumed. The new framework is evaluated in a virtual environment under different environmental and model settings, using random simulations of more than 165,000 swarm flights. For certain environmental settings, our results indicate that the interception performance of larger swarms under partial observation is comparable to that of a one-drone swarm under full observation of the adversarial drone.
{"title":"Interception of automated adversarial drone swarms in partially observed environments","authors":"Daniel Saranovic, M. Pavlovski, W. Power, Ivan Stojkovic, Z. Obradovic","doi":"10.3233/ICA-210653","DOIUrl":"https://doi.org/10.3233/ICA-210653","url":null,"abstract":"As the prevalence of drones increases, understanding and preparing for possible adversarial uses of drones and drone swarms is of paramount importance. Correspondingly, developing defensive mechanisms in which swarms can be used to protect against adversarial Unmanned Aerial Vehicles (UAVs) is a problem that requires further attention. Prior work on intercepting UAVs relies mostly on utilizing additional sensors or uses the Hamilton-Jacobi-Bellman equation, for which strong conditions need to be met to guarantee the existence of a saddle-point solution. To that end, this work proposes a novel interception method that utilizes the swarm’s onboard PID controllers for setting the drones’ states during interception. The drone’s states are constrained only by their physical limitations, and only partial feedback of the adversarial drone’s positions is assumed. The new framework is evaluated in a virtual environment under different environmental and model settings, using random simulations of more than 165,000 swarm flights. For certain environmental settings, our results indicate that the interception performance of larger swarms under partial observation is comparable to that of a one-drone swarm under full observation of the adversarial drone.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-04-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90262750","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In technical systems the analysis of similar situations is a promising technique to gain information about the system’s state, its health or wearing. Very often, situations cannot be defined but need to be discovered as recurrent patterns within time series data of the system under consideration. This paper addresses the assessment of different approaches to discover frequent variable-length patterns in time series. Because of the success of artificial neural networks (NN) in various research fields, a special issue of this work is the applicability of NNs to the problem of pattern discovery in time series. Therefore we applied and adapted a Convolutional Autoencoder and compared it to classical nonlearning approaches based on Dynamic Time Warping, based on time series discretization as well as based on the Matrix Profile. These nonlearning approaches have also been adapted, to fulfill our requirements like the discovery of potentially time scaled patterns from noisy time series. We showed the performance (quality, computing time, effort of parametrization) of those approaches in an extensive test with synthetic data sets. Additionally the transferability to other data sets is tested by using real life vehicle data. We demonstrated the ability of Convolutional Autoencoders to discover patterns in an unsupervised way. Furthermore the tests showed, that the Autoencoder is able to discover patterns with a similar quality like classical nonlearning approaches.
{"title":"Pattern discovery in time series using autoencoder in comparison to nonlearning approaches","authors":"F. Noering, Yannik Schröder, K. Jonas, F. Klawonn","doi":"10.3233/ICA-210650","DOIUrl":"https://doi.org/10.3233/ICA-210650","url":null,"abstract":"In technical systems the analysis of similar situations is a promising technique to gain information about the system’s state, its health or wearing. Very often, situations cannot be defined but need to be discovered as recurrent patterns within time series data of the system under consideration. This paper addresses the assessment of different approaches to discover frequent variable-length patterns in time series. Because of the success of artificial neural networks (NN) in various research fields, a special issue of this work is the applicability of NNs to the problem of pattern discovery in time series. Therefore we applied and adapted a Convolutional Autoencoder and compared it to classical nonlearning approaches based on Dynamic Time Warping, based on time series discretization as well as based on the Matrix Profile. These nonlearning approaches have also been adapted, to fulfill our requirements like the discovery of potentially time scaled patterns from noisy time series. We showed the performance (quality, computing time, effort of parametrization) of those approaches in an extensive test with synthetic data sets. Additionally the transferability to other data sets is tested by using real life vehicle data. We demonstrated the ability of Convolutional Autoencoders to discover patterns in an unsupervised way. Furthermore the tests showed, that the Autoencoder is able to discover patterns with a similar quality like classical nonlearning approaches.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81777766","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A Must-read Journal for Engineering","authors":"A. D. L. Escalera","doi":"10.3233/ICA-210658","DOIUrl":"https://doi.org/10.3233/ICA-210658","url":null,"abstract":"","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"86018562","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jan Ga̧sienica-Józkowy, Mateusz Knapik, B. Cyganek
Today’s deep learning architectures, if trained with proper dataset, can be used for object detection in marine search and rescue operations. In this paper a dataset for maritime search and rescue purposes is proposed. It contains aerial-drone videos with 40,000 hand-annotated persons and objects floating in the water, many of small size, which makes them difficult to detect. The second contribution is our proposed object detection method. It is an ensemble composed of a number of the deep convolutional neural networks, orchestrated by the fusion module with the nonlinearly optimized voting weights. The method achieves over 82% of average precision on the new aerial-drone floating objects dataset and outperforms each of the state-of-the-art deep neural networks, such as YOLOv3, -v4, Faster R-CNN, RetinaNet, and SSD300. The dataset is publicly available from the Internet.
{"title":"An ensemble deep learning method with optimized weights for drone-based water rescue and surveillance","authors":"Jan Ga̧sienica-Józkowy, Mateusz Knapik, B. Cyganek","doi":"10.3233/ICA-210649","DOIUrl":"https://doi.org/10.3233/ICA-210649","url":null,"abstract":"Today’s deep learning architectures, if trained with proper dataset, can be used for object detection in marine search and rescue operations. In this paper a dataset for maritime search and rescue purposes is proposed. It contains aerial-drone videos with 40,000 hand-annotated persons and objects floating in the water, many of small size, which makes them difficult to detect. The second contribution is our proposed object detection method. It is an ensemble composed of a number of the deep convolutional neural networks, orchestrated by the fusion module with the nonlinearly optimized voting weights. The method achieves over 82% of average precision on the new aerial-drone floating objects dataset and outperforms each of the state-of-the-art deep neural networks, such as YOLOv3, -v4, Faster R-CNN, RetinaNet, and SSD300. The dataset is publicly available from the Internet.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90223169","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
M. J. Gómez-Silva, A. D. L. Escalera, J. M. Armingol
The automatization of the Re-Identification of an individual across different video-surveillance cameras poses a significant challenge due to the presence of a vast number of potential candidates with a similar appearance. This task requires the learning of discriminative features from person images and a distance metric to properly compare them and decide whether they belong to the same person or not. Nevertheless, the fact of acquiring images of the same person from different, distant and non-overlapping views produces changes in illumination, perspective, background, resolution and scale between the person’s representations, resulting in appearance variations that hamper his/her re-identification. This article focuses the feature learning on automatically finding discriminative descriptors able to reflect the dissimilarities mainly due to the changes in actual people appearance, independently from the variations introduced by the acquisition point. With that purpose, such variations have been implicitly embedded by the Mahalanobis distance. This article presents a learning algorithm to jointly model features and the Mahalanobis distance through a Deep Neural Re-Identification model. The Mahalanobis distance learning has been implemented as a novel neural layer, forming part of a Triplet Learning model that has been evaluated over PRID2011 dataset, providing satisfactory results.
{"title":"Back-propagation of the Mahalanobis istance through a deep triplet learning model for person Re-Identification","authors":"M. J. Gómez-Silva, A. D. L. Escalera, J. M. Armingol","doi":"10.3233/ICA-210651","DOIUrl":"https://doi.org/10.3233/ICA-210651","url":null,"abstract":"The automatization of the Re-Identification of an individual across different video-surveillance cameras poses a significant challenge due to the presence of a vast number of potential candidates with a similar appearance. This task requires the learning of discriminative features from person images and a distance metric to properly compare them and decide whether they belong to the same person or not. Nevertheless, the fact of acquiring images of the same person from different, distant and non-overlapping views produces changes in illumination, perspective, background, resolution and scale between the person’s representations, resulting in appearance variations that hamper his/her re-identification. This article focuses the feature learning on automatically finding discriminative descriptors able to reflect the dissimilarities mainly due to the changes in actual people appearance, independently from the variations introduced by the acquisition point. With that purpose, such variations have been implicitly embedded by the Mahalanobis distance. This article presents a learning algorithm to jointly model features and the Mahalanobis distance through a Deep Neural Re-Identification model. The Mahalanobis distance learning has been implemented as a novel neural layer, forming part of a Triplet Learning model that has been evaluated over PRID2011 dataset, providing satisfactory results.","PeriodicalId":50358,"journal":{"name":"Integrated Computer-Aided Engineering","volume":null,"pages":null},"PeriodicalIF":6.5,"publicationDate":"2021-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74690035","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：