Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981586
M. Santos, G. G. Giacomo, Paulo L. J. Drews-Jr, S. Botelho
Autonomous underwater navigation is a challenging problem because of the limitations imposed by aquatic environments. Among them, the use of Global Positioning System (GPS) is severely limited. Thus, we propose the use of sensor fusion to improve underwater localization in partially structured environments. We sustain our proposal explores the benefits of aerial images, such as georeferencing, to improve underwater navigation with a multibeam forward looking sonar. Our methodology combines state-of-the-art approaches such as Deep Neural Networks and Adaptive Monte Carlo Localization to fuse data from different image domains. The obtained results show a significant improvement over traditional odometry for underwater localization.
{"title":"Underwater Sonar and Aerial Images Data Fusion for Robot Localization","authors":"M. Santos, G. G. Giacomo, Paulo L. J. Drews-Jr, S. Botelho","doi":"10.1109/ICAR46387.2019.8981586","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981586","url":null,"abstract":"Autonomous underwater navigation is a challenging problem because of the limitations imposed by aquatic environments. Among them, the use of Global Positioning System (GPS) is severely limited. Thus, we propose the use of sensor fusion to improve underwater localization in partially structured environments. We sustain our proposal explores the benefits of aerial images, such as georeferencing, to improve underwater navigation with a multibeam forward looking sonar. Our methodology combines state-of-the-art approaches such as Deep Neural Networks and Adaptive Monte Carlo Localization to fuse data from different image domains. The obtained results show a significant improvement over traditional odometry for underwater localization.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"45 1","pages":"578-583"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73688544","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981654
Silas Grün, Simon Höninger, Paul Scheikl, B. Hein, T. Kröger
To address the challenge of resource-intensive data collection from real robotic environments, many deep learning applications use synthetic data to train their networks. This creates new problems when transferring the obtained knowledge from the simulated to the real world domain. Various aspects of the simulation, which do not influence the learning objective, can be randomized to enhance generalization to new domains. In this paper, we analyze the effect of these domain randomization techniques. To get an insight into their benefits, we apply them while training a grasp success classifier based on state-of-the-art CNN for an industrial robot as a showcase. We generated a large synthetic data set containing 1.44M RGB images with 48 permutations of 6 different randomizations and a base scenario as training data. The resulting networks, each trained on a different subset of this data set, are evaluated on 3k real world images of the robot performing grasps. We observed the effectiveness of randomization of perspective, distractors, lighting and the grasped box. Notably, we show that pretrained networks benefit from these techniques in particular.
{"title":"Evaluation of Domain Randomization Techniques for Transfer Learning","authors":"Silas Grün, Simon Höninger, Paul Scheikl, B. Hein, T. Kröger","doi":"10.1109/ICAR46387.2019.8981654","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981654","url":null,"abstract":"To address the challenge of resource-intensive data collection from real robotic environments, many deep learning applications use synthetic data to train their networks. This creates new problems when transferring the obtained knowledge from the simulated to the real world domain. Various aspects of the simulation, which do not influence the learning objective, can be randomized to enhance generalization to new domains. In this paper, we analyze the effect of these domain randomization techniques. To get an insight into their benefits, we apply them while training a grasp success classifier based on state-of-the-art CNN for an industrial robot as a showcase. We generated a large synthetic data set containing 1.44M RGB images with 48 permutations of 6 different randomizations and a base scenario as training data. The resulting networks, each trained on a different subset of this data set, are evaluated on 3k real world images of the robot performing grasps. We observed the effectiveness of randomization of perspective, distractors, lighting and the grasped box. Notably, we show that pretrained networks benefit from these techniques in particular.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"17 1","pages":"481-486"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72639185","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981566
Shenghao Jiang, Macheng Shen
We propose a novel cognitive collision avoidance (CA) framework for autonomous driving (AD) vehicles in urban environments. In this framework, a hybrid future trajectory predictor is developed, which consists of a static agent classifier, a recurrent neural network (RNN) based trajectory predictor and a lane-based kinematic model predictor. To fuse the outputs of different predictors, an iterative multivariate Gaussian weighted algorithm is designed to drop outliers and estimate the predicted dynamic features more reliably. Subsequently, fed in with the fused results of observed agents, together with the current dynamic features and planned trajectory of the ego vehicle, an RNN-based priority prediction engine is applied to infer the priority probabilities distribution for CA decision, which indicates the likelihood that the vehicle continue driving according to its planned trajectory. By observing surrounding agents' historical ground truth trajectory and taking the road geometry constraints into consideration, the future dynamic features, priority probabilities distribution and the CA decision can be figured out at every timestamp cognitively and adaptively. The performance of this framework is evaluated on a prototype car in multiple typical USA urban scenarios, comparing with conventional CA systems which assume constant velocity and only work when observed agents follow traffic rules, our framework alleviates these limitations and achieves encouraging results in terms of the priority distribution estimation, with a frequency >20Hz, which is capable of running in real-time.
{"title":"A Cognitive Urban Collision Avoidance Framework Based on Agents Priority Using Recurrent Neural Network","authors":"Shenghao Jiang, Macheng Shen","doi":"10.1109/ICAR46387.2019.8981566","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981566","url":null,"abstract":"We propose a novel cognitive collision avoidance (CA) framework for autonomous driving (AD) vehicles in urban environments. In this framework, a hybrid future trajectory predictor is developed, which consists of a static agent classifier, a recurrent neural network (RNN) based trajectory predictor and a lane-based kinematic model predictor. To fuse the outputs of different predictors, an iterative multivariate Gaussian weighted algorithm is designed to drop outliers and estimate the predicted dynamic features more reliably. Subsequently, fed in with the fused results of observed agents, together with the current dynamic features and planned trajectory of the ego vehicle, an RNN-based priority prediction engine is applied to infer the priority probabilities distribution for CA decision, which indicates the likelihood that the vehicle continue driving according to its planned trajectory. By observing surrounding agents' historical ground truth trajectory and taking the road geometry constraints into consideration, the future dynamic features, priority probabilities distribution and the CA decision can be figured out at every timestamp cognitively and adaptively. The performance of this framework is evaluated on a prototype car in multiple typical USA urban scenarios, comparing with conventional CA systems which assume constant velocity and only work when observed agents follow traffic rules, our framework alleviates these limitations and achieves encouraging results in terms of the priority distribution estimation, with a frequency >20Hz, which is capable of running in real-time.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"119 1","pages":"474-480"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80393832","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981600
Jens Lambrecht, Linh Kästner
Pose estimation is a necessity for many applications in robotics incorporating interaction between the robot and external camera-equipped devices, e.g. mobile robots or Augmented Reality devices. In the practice of monocular cameras, one mostly takes advantage of pose estimation through fiducial marker detection. We propose a novel approach for marker-less robot pose estimation through monocular cameras utilizing 2D keypoint detection and 3D keypoint determination through readings from the encoders and forward kinematics. In particular, 2D-3D point correspondences enable the pose estimation through solving the Perspective-n-Point problem for calibrated cameras. The method does not rely on any depth data or initializations. The robust 2D keypoint detection is implemented by modern Convolutional Neural Networks trained on different dataset configurations of real and synthetic data in order to quantitatively evaluate robustness, precision and data efficiency. We demonstrate that the method provides robust pose estimation for random joint poses and benchmark the performance of different (synthetic) dataset configurations. Furthermore, we compare the accuracies to marker pose estimation and give an outlook towards enhancements and realtime capability.
{"title":"Towards the Usage of Synthetic Data for Marker-Less Pose Estimation of Articulated Robots in RGB Images","authors":"Jens Lambrecht, Linh Kästner","doi":"10.1109/ICAR46387.2019.8981600","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981600","url":null,"abstract":"Pose estimation is a necessity for many applications in robotics incorporating interaction between the robot and external camera-equipped devices, e.g. mobile robots or Augmented Reality devices. In the practice of monocular cameras, one mostly takes advantage of pose estimation through fiducial marker detection. We propose a novel approach for marker-less robot pose estimation through monocular cameras utilizing 2D keypoint detection and 3D keypoint determination through readings from the encoders and forward kinematics. In particular, 2D-3D point correspondences enable the pose estimation through solving the Perspective-n-Point problem for calibrated cameras. The method does not rely on any depth data or initializations. The robust 2D keypoint detection is implemented by modern Convolutional Neural Networks trained on different dataset configurations of real and synthetic data in order to quantitatively evaluate robustness, precision and data efficiency. We demonstrate that the method provides robust pose estimation for random joint poses and benchmark the performance of different (synthetic) dataset configurations. Furthermore, we compare the accuracies to marker pose estimation and give an outlook towards enhancements and realtime capability.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"62 1","pages":"240-247"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81728039","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981559
D. K. Villa, A. Brandão, M. S. Filho
This paper addresses the problem of carrying a rod-shaped load between a certain origin and a desired goal using unmanned aerial vehicles (UAV). The load is carried via flexible cables by two quadrotors, one at each end of the bar. Positioning, orientation, and path-following tasks are here addressed. The robots and the load are modeled as a single system, using a virtual structure framework for robot formation and a nonlinear controller based on feedback linearization to handle the load oscillations and accomplish the missions. Results obtained running a real experiment using two AR. Drone quadrotors to carry an aluminum bar are presented through illustrations and videos, which validate the proposed system.
{"title":"Path-Following and Attitude Control of a Payload Using Multiple Quadrotors","authors":"D. K. Villa, A. Brandão, M. S. Filho","doi":"10.1109/ICAR46387.2019.8981559","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981559","url":null,"abstract":"This paper addresses the problem of carrying a rod-shaped load between a certain origin and a desired goal using unmanned aerial vehicles (UAV). The load is carried via flexible cables by two quadrotors, one at each end of the bar. Positioning, orientation, and path-following tasks are here addressed. The robots and the load are modeled as a single system, using a virtual structure framework for robot formation and a nonlinear controller based on feedback linearization to handle the load oscillations and accomplish the missions. Results obtained running a real experiment using two AR. Drone quadrotors to carry an aluminum bar are presented through illustrations and videos, which validate the proposed system.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"36 1","pages":"535-540"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84343888","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981571
Ignacio J. Sánchez, A. Ferramosca, G. Raffo, A. González, A. D'Jorge
This work presents a model predictive formulation for obstacle avoiding path following control for constrained vehicles. The obstacles are introduced as soft constraints in the value function, in order to maintain the convexity of state and output spaces. In this formulation, the path following and obstacle avoidance tasks may introduce local minima solutions -due to their competing costs- known as corner conditions. In order to address this problem, a heuristic switch in the form of additional decision variables is introduced into the cost function. The proposed solution is based on an extension of Model Predictive Control (MPC) by using Artificial Variables. An additional cost term is included in order to prevent early stops in the path following task. Simulations results considering an autonomous vehicle subject to input constraints are carried out to illustrate the performance of the proposed control strategy.
{"title":"Obstacle Avoiding Path Following based on Nonlinear Model Predictive Control using Artificial Variables","authors":"Ignacio J. Sánchez, A. Ferramosca, G. Raffo, A. González, A. D'Jorge","doi":"10.1109/ICAR46387.2019.8981571","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981571","url":null,"abstract":"This work presents a model predictive formulation for obstacle avoiding path following control for constrained vehicles. The obstacles are introduced as soft constraints in the value function, in order to maintain the convexity of state and output spaces. In this formulation, the path following and obstacle avoidance tasks may introduce local minima solutions -due to their competing costs- known as corner conditions. In order to address this problem, a heuristic switch in the form of additional decision variables is introduced into the cost function. The proposed solution is based on an extension of Model Predictive Control (MPC) by using Artificial Variables. An additional cost term is included in order to prevent early stops in the path following task. Simulations results considering an autonomous vehicle subject to input constraints are carried out to illustrate the performance of the proposed control strategy.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"48 1","pages":"254-259"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82041673","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981581
Sebastian Rendon Fernandez, A. Olabi, O. Gibaru
The robotization of assembly operations is one of the requests of aircraft manufacturers. During the assembly of large-scale sub-assemblies, contact forces between parts must be controlled. Exceeding some limits can damage the aircraft parts. This can happen because of the poor accuracy of industrial robots and uncertainties of parts' positions, and orientations. This paper proposes a new approach to control the movement of the robot end-effector, taking into account the contact forces between the parts during assembly. The suggested approach can be used to assemble complex shape and large-scale parts. Based on the admittance control, the proposed approach is used to ensure multi-surface contact. It allows to control the interaction forces at each contact surface. Each contact is modeled by a mass-spring-damper system. This approach was tested on the assembly of two large-scale airplane's parts using a KUKA robot (KR340), equipped with a Force/Torque (F/T) sensor. The performance of this multi-surface approach was compared to one surface admittance control.
{"title":"Multi-Surface Admittance Control Approach applied on Robotic Assembly of Large-Scale parts in Aerospace Manufacturing","authors":"Sebastian Rendon Fernandez, A. Olabi, O. Gibaru","doi":"10.1109/ICAR46387.2019.8981581","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981581","url":null,"abstract":"The robotization of assembly operations is one of the requests of aircraft manufacturers. During the assembly of large-scale sub-assemblies, contact forces between parts must be controlled. Exceeding some limits can damage the aircraft parts. This can happen because of the poor accuracy of industrial robots and uncertainties of parts' positions, and orientations. This paper proposes a new approach to control the movement of the robot end-effector, taking into account the contact forces between the parts during assembly. The suggested approach can be used to assemble complex shape and large-scale parts. Based on the admittance control, the proposed approach is used to ensure multi-surface contact. It allows to control the interaction forces at each contact surface. Each contact is modeled by a mass-spring-damper system. This approach was tested on the assembly of two large-scale airplane's parts using a KUKA robot (KR340), equipped with a Force/Torque (F/T) sensor. The performance of this multi-surface approach was compared to one surface admittance control.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"54 1","pages":"688-694"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85790736","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981663
Gabriele Bolano, Pascal Becker, Jacques Kaiser, A. Roennau, R. Dillmann
Service robots are becoming able to perform a variety of tasks and they are currently used for many different applications. For this reason people with different backgrounds and also without robotic experience need to interact with them. Enabling the user to control the motion of the robot end-effector, it is important to provide an easy and intuitive interface. In this work we propose an intuitive method for the control of a robot TCP position and orientation. This is done taking into account the robot kinematics in order to avoid dangerous configuration and defining rotational constraints. The user is enabled to interact with the robot and control its end-effector using a set of objects tracked by a camera system. The autonomy level of the robot changes depending on the different phases of the interaction for a better efficiency. An intuitive GUI has been developed to ease the interaction and help the user to achieve a better precision in the control. This is possible also through the scaling of the tracked motion, which is represented as visual feedback. We tested the system through multiple experiments that took into account how people with no experience interact with the robot and the precision of the method.
{"title":"Advanced Usability Through Constrained Multi Modal Interactive Strategies: The CookieBot","authors":"Gabriele Bolano, Pascal Becker, Jacques Kaiser, A. Roennau, R. Dillmann","doi":"10.1109/ICAR46387.2019.8981663","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981663","url":null,"abstract":"Service robots are becoming able to perform a variety of tasks and they are currently used for many different applications. For this reason people with different backgrounds and also without robotic experience need to interact with them. Enabling the user to control the motion of the robot end-effector, it is important to provide an easy and intuitive interface. In this work we propose an intuitive method for the control of a robot TCP position and orientation. This is done taking into account the robot kinematics in order to avoid dangerous configuration and defining rotational constraints. The user is enabled to interact with the robot and control its end-effector using a set of objects tracked by a camera system. The autonomy level of the robot changes depending on the different phases of the interaction for a better efficiency. An intuitive GUI has been developed to ease the interaction and help the user to achieve a better precision in the control. This is possible also through the scaling of the tracked motion, which is represented as visual feedback. We tested the system through multiple experiments that took into account how people with no experience interact with the robot and the precision of the method.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"260 1","pages":"213-219"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77931134","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981608
G. Tejera, G. Amorín, Andrés Seré, Nicolás Capricho, Pablo Margenat, J. Visca
Computational thinking is a skill that is considered essential for the future generations. Because of this it should be incorporated into the curricula as soon as possible. Many robots can be programmed using graphical languages or physical blocks instead of writing code. This makes programming more accessible for the youngest programmers. Looking to extend the programming activities to preschool students, we present a novel approach that allows to program a mobile robot, Robotito, by changing its environment. We describe the architecture of Robotito, software used to program its interaction with the environment, and developed behaviours. Moreover, Robotito exports his sensors and actuators using ROS standard mechanisms and is modelled in Gazebo allowing it to be used in research and undergraduate school courses providing researchers an autonomous and safety mobile platform, which can be integrated with any system using ROS.
{"title":"Robotito: programming robots from preschool to undergraduate school level","authors":"G. Tejera, G. Amorín, Andrés Seré, Nicolás Capricho, Pablo Margenat, J. Visca","doi":"10.1109/ICAR46387.2019.8981608","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981608","url":null,"abstract":"Computational thinking is a skill that is considered essential for the future generations. Because of this it should be incorporated into the curricula as soon as possible. Many robots can be programmed using graphical languages or physical blocks instead of writing code. This makes programming more accessible for the youngest programmers. Looking to extend the programming activities to preschool students, we present a novel approach that allows to program a mobile robot, Robotito, by changing its environment. We describe the architecture of Robotito, software used to program its interaction with the environment, and developed behaviours. Moreover, Robotito exports his sensors and actuators using ROS standard mechanisms and is modelled in Gazebo allowing it to be used in research and undergraduate school courses providing researchers an autonomous and safety mobile platform, which can be integrated with any system using ROS.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"216 1","pages":"296-301"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"72875052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-12-01DOI: 10.1109/ICAR46387.2019.8981569
Lea Steffen, Stefan Ulbrich, A. Roennau, R. Dillmann
Depth perception is crucial for many applications including robotics, UAV and autonomous driving. The visual sense, as well as cameras, map the 3D world on a 2D representation, losing the dimension representing depth. A way to recover 3D information from 2D images is to record and join data from multiple viewpoints. In case of a stereo setup, 4D data is gained. Existing methods to recover 3D information are computationally expensive. We propose a new, more intuitive method to recover 3D objects out of event-based stereo data, by using a Self-Organizing Map to solve the correspondence problem and establish a structure similar to a voxel grid. Our approach, as it is also computationally expensive, copes with performance issues by massive parallelization. Furthermore, the relatively small voxel grid makes this a memory friendly solution. This technique is very powerful as it does not need any prior knowledge of extrinsic and intrinsic camera parameters. Instead, those parameters and also the lens distortion are learned implicitly. Not only do we not require a parallel camera setup, as many existing methods, we do not even need any information about the alignment at all. We evaluated our method in a qualitative analysis and finding image correspondences.
{"title":"Multi-View 3D Reconstruction with Self-Organizing Maps on Event-Based Data","authors":"Lea Steffen, Stefan Ulbrich, A. Roennau, R. Dillmann","doi":"10.1109/ICAR46387.2019.8981569","DOIUrl":"https://doi.org/10.1109/ICAR46387.2019.8981569","url":null,"abstract":"Depth perception is crucial for many applications including robotics, UAV and autonomous driving. The visual sense, as well as cameras, map the 3D world on a 2D representation, losing the dimension representing depth. A way to recover 3D information from 2D images is to record and join data from multiple viewpoints. In case of a stereo setup, 4D data is gained. Existing methods to recover 3D information are computationally expensive. We propose a new, more intuitive method to recover 3D objects out of event-based stereo data, by using a Self-Organizing Map to solve the correspondence problem and establish a structure similar to a voxel grid. Our approach, as it is also computationally expensive, copes with performance issues by massive parallelization. Furthermore, the relatively small voxel grid makes this a memory friendly solution. This technique is very powerful as it does not need any prior knowledge of extrinsic and intrinsic camera parameters. Instead, those parameters and also the lens distortion are learned implicitly. Not only do we not require a parallel camera setup, as many existing methods, we do not even need any information about the alignment at all. We evaluated our method in a qualitative analysis and finding image correspondences.","PeriodicalId":6606,"journal":{"name":"2019 19th International Conference on Advanced Robotics (ICAR)","volume":"110 1","pages":"501-508"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76760634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","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：