Article
Computer Science, Artificial Intelligence
Qiyu Sun, Yang Tang, Chongzhen Zhang, Chaoqiang Zhao, Feng Qian, Jurgen Kurths
Summary: In this work, the negative impact of dynamic environments on the joint estimation of depth and visual odometry (VO) is alleviated through hybrid masks. The proposed cover mask and filter mask help to improve the accuracy of VO estimation and depth estimation in the presence of dynamic environments.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Chemistry, Analytical
Claudio Cimarelli, Hriday Bavle, Jose Luis Sanchez-Lopez, Holger Voos
Summary: This paper presents a method for frame-to-frame motion estimation and rotation adjustment based on a model-free epipolar constraint. By matching 2D keypoints using pre-trained deep networks and utilizing an unsupervised training protocol, the proposed method achieves considerable improvements in accuracy and complexity compared to other unsupervised pose networks.
Article
Computer Science, Artificial Intelligence
Kutsev Bengisu Ozyoruk, Guliz Irem Gokceler, Taylor L. Bobrow, Gulfize Coskun, Kagan Incetan, Yasin Almalioglu, Faisal Mahmood, Eva Curto, Luis Perdigoto, Marina Oliveira, Hasan Sahin, Helder Araujo, Henrique Alexandrino, Nicholas J. Durr, Hunter B. Gilbert, Mehmet Turan
Summary: Deep learning techniques show promise in developing dense topography reconstruction and pose estimation methods for endoscopic videos, but lack of effective quantitative benchmarking datasets. The introduced comprehensive endoscopic SLAM dataset includes 3D point cloud data, capsule and standard endoscopy recordings, synthetically generated data, and clinically used endoscope recordings. The dataset is validated for real clinical systems and includes synthetic capsule endoscopy frames with annotations for simulation-to-real transfer learning algorithms.
MEDICAL IMAGE ANALYSIS
(2021)
Article
Chemistry, Analytical
Henghui Zhi, Chenyang Yin, Huibin Li, Shanmin Pang
Summary: This article introduces an unsupervised monocular visual odometry framework that can model multi-scale information to improve the accuracy of pose and depth estimation in rotating scenes.
Article
Chemistry, Analytical
Xudong Zhang, Baigan Zhao, Jiannan Yao, Guoqing Wu
Summary: This paper presents a novel unsupervised learning framework for estimating scene depth and camera pose from video sequences. Multiple mask technologies and geometric consistency constraints are employed to mitigate the negative impacts of challenging scenes, such as dynamic objects and occluded regions. Experimental results on the KITTI dataset demonstrate that these strategies effectively enhance the model's performance, outperforming other unsupervised methods.
Article
Chemistry, Analytical
Yuji Zhuang, Xiaoyan Jiang, Yongbin Gao, Zhijun Fang, Hamido Fujita
Summary: This paper investigates the importance of robust and accurate visual feature tracking for pose estimation in visual odometry. It proposes an unsupervised monocular visual odometry framework that combines features extracted from two sources, an optical flow network and a traditional point feature extractor. Experimental results demonstrate the robust performance of the proposed method in complicated fast-motion scenarios.
Article
Engineering, Electrical & Electronic
Lili Lin, Weisheng Wang, Wan Luo, Lesheng Song, Wenhui Zhou
Summary: This paper introduces a novel visual odometry method that accurately estimates the camera pose by separating pose estimation and refinement, in order to address the issue of drift or error accumulation in visual odometry.
DIGITAL SIGNAL PROCESSING
(2021)
Article
Robotics
Libo Sun, Wei Yin, Enze Xie, Zhengrong Li, Changming Sun, Chunhua Shen
Summary: In this article, we propose a framework to improve monocular visual odometry (VO) systems by utilizing monocular depth estimation. Our framework has a strong generalization capability and can improve the accuracy of localization and mapping. Compared to current learning-based methods, our method demonstrates better adaptability to diverse scenes and significantly boosts the performance of geometry-based methods.
IEEE TRANSACTIONS ON ROBOTICS
(2022)
Article
Engineering, Civil
Yiling Liu, Hesheng Wang, Jingchuan Wang, Xinlei Wang
Summary: This paper proposes an end-to-end unsupervised visual odometry framework based on deep learning, which improves the accuracy and robustness of pose estimation through confidence evaluation.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Automation & Control Systems
Haixin Xiu, Yiyou Liang, Hui Zeng, Qing Li, Hongmin Liu, Bin Fan, Chen Li
Summary: This paper proposes a novel prediction-update pose estimation network, PU-PoseNet, for self-supervised monocular visual odometry. It enhances the time consistency and robustness of estimation results through long-time pose consistency constraint, depth consistency constraint, and automatic masking. Furthermore, a frame missing training strategy is used to adapt to missing frames.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Automation & Control Systems
Ling Li, Xiaojian Li, Shanlin Yang, Shuai Ding, Alireza Jolfaei, Xi Zheng
Summary: This study proposed a fully unsupervised learning method for depth and motion estimation using continuous monocular endoscopic video. By designing EndoMotionNet and EndoDepthNet models, considering timing information, and implementing a multimode fusion mechanism, the accuracy of depth and motion estimation can be significantly improved.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Engineering, Electrical & Electronic
Guangming Wang, Jiquan Zhong, Shijie Zhao, Wenhua Wu, Zhe Liu, Hesheng Wang
Summary: In this paper, a novel unsupervised training framework is proposed to learn per-pixel depth and ego-motion from unlabeled monocular video. The framework utilizes 3D hierarchical refinement and augmentation using explicit 3D geometry to iteratively optimize depth and pose estimations. The proposed method achieves state-of-the-art performance in depth estimation and outperforms recent unsupervised monocular learning-based methods in visual odometry.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Civil
Shaocheng Jia, Xin Pei, Xiao Jing, Danya Yao
Summary: This paper proposes a novel scheme of correlation-aware structure to explore relations between depths and utilizes a Gaussian estimator to predict depth map and uncertainty map simultaneously. Strategies based on uncertainty are developed to improve performance, and experiments show competitive results compared to state-of-the-art methods. The method demonstrates strong generalization capability and practicality in additional experiments on different datasets.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Saul Martinez-Diaz
Summary: This paper introduces a new method for estimating the distance of an object from a single image using a monocular calibrated vision system combined with a Region-based Convolutional Neural Network. Experimental results showed that the proposed method accomplished the task in less time and exhibited similar performance to a calibrated stereo vision system.
JOURNAL OF SENSORS
(2021)
Article
Computer Science, Artificial Intelligence
Zhongyi Wang, Mengjiao Shen, Qijun Chen
Summary: Scale ambiguity is a challenge in monocular visual odometry, which has not been well addressed by traditional or learning-based methods. Unsupervised monocular visual odometry also has limited performance in tricky situations. To overcome these issues, we propose an accurate and efficient end-to-end system that can effectively handle scale ambiguity in unsupervised monocular visual odometry, especially in challenging long-sequence videos. By utilizing depth and optical flow networks, we obtain the depth map and optical flow information. The combination of optical flow and camera height, along with our scale recovery algorithm, allows us to recover the absolute scale of camera translation and depth map. Extensive experiments show that our method achieves state-of-the-art performance on the KITTI dataset for depth and optical flow estimation, as well as monocular visual odometry.
NEURAL PROCESSING LETTERS
(2023)
Article
Chemistry, Physical
Jacob Johny, Oleg Prymak, Marius Kamp, Florent Calvo, Se-Ho Kim, Anna Tymoczko, Ayman El-Zoka, Christoph Rehbock, Ulrich Schuermann, Baptiste Gault, Lorenz Kienle, Stephan Barcikowski
Summary: Bimetallic nanoparticles are promising for various applications due to their enhanced properties, with the Au-Fe system showing potential for catalysis. The study reveals the complex evolution of crystal structures in nanoparticles at high temperatures, highlighting the importance of understanding such changes for future applications in high-temperature processes.
Article
Chemistry, Multidisciplinary
Se-Ho Kim, Su-Hyun Yoo, Poulami Chakraborty, Jiwon Jeong, Joohyun Lim, Ayman A. El-Zoka, Xuyang Zhou, Leigh T. Stephenson, Tilmann Hickel, Joerg Neugebauer, Christina Scheu, Mira Todorova, Baptiste Gault
Summary: Metal nanogels possess a large surface area, high structural stability, and high catalytic activity, which are determined by the atomic-level distribution of their constituents. However, analyzing their subnanoscale structure and composition for property optimization is challenging. In this study, Pd nanogels were synthesized and an analysis revealed that impurities from the reactants integrated into the grain boundaries, which are typically sites of high catalytic activity. The level of impurities was found to be controlled by the reaction conditions, offering opportunities for designing new nanogels.
JOURNAL OF THE AMERICAN CHEMICAL SOCIETY
(2022)
Article
Chemistry, Physical
Se-Ho Kim, Kang Dong, Huan Zhao, Ayman A. El-Zoka, Xuyang Zhou, Eric V. Woods, Finn Giuliani, Ingo Manke, Dierk Raabe, Baptiste Gault
Summary: By using cryo-atom probe tomography, we have discovered the degradation process between the liquid electrolyte and Si electrode. We found that the Si electrode corrodes before the charge-discharge cycles begin, and the delithiation process leads to the formation of nanograins. These nanograins are pulverized into nanoscale fragments that float in the electrolyte. The electrolyte also decomposes during this process. Understanding these microstructures is crucial for understanding the degradation of Si anodes and can potentially inform the design of new batteries.
JOURNAL OF PHYSICAL CHEMISTRY LETTERS
(2022)
Article
Materials Science, Multidisciplinary
Chanwon Jung, Hosun Jun, Kyuseon Jang, Se-Ho Kim, Pyuck-Pa Choi
Summary: This study utilizes atom probe tomography to investigate the structure-property relationships of carbon-supported nanoparticles. It synthesizes and analyzes carbon-supported Pt, PtMn alloy, and ordered Pt3Mn nanoparticles as model systems. The research provides insights into the 3D elemental distribution and the field evaporation behavior of the carbon support, and offers guidance for future studies and applications.
MICROSCOPY AND MICROANALYSIS
(2022)
Article
Materials Science, Multidisciplinary
Se-Ho Kim, Leigh T. Stephenson, Alisson K. da Silva, Baptiste Gault, Ayman A. El-Zoka
Summary: This study investigates the phase separation and mixing processes of EGaIn alloy during thermal cycling using cryogenically-enabled advanced microscopy and microanalysis. The results reveal a thermal-stimulus-response behavior of EGaIn at cryogenic temperatures, with a sudden volume expansion associated with the formation of metastable Ga phases. These findings highlight the importance of rejuvenation kinetics and open new possibilities for the application of EGaIn as a self-healing material.
Article
Materials Science, Multidisciplinary
Se-Ho Kim, Leigh T. Stephenson, Torsten Schwarz, Baptiste Gault
Summary: This study compared the distribution of constituent elements in two different smartphone glass samples using X-ray spectroscopy techniques and atom probe tomography (APT). The results demonstrated that APT can be considered as an alternative technique for imaging the chemical distribution in glass for mobile applications.
MICROSCOPY AND MICROANALYSIS
(2023)
Article
Physics, Multidisciplinary
Xuyang Zhou, Yang Bai, Ayman A. El-Zoka, Se -Ho Kim, Yan Ma, Christian H. Liebscher, Baptiste Gault, Jaber R. Mianroodi, Gerhard Dehm, Dierk Raabe
Summary: Solid-state redox-driven phase transformations can create pores, and the accumulation of redox products inside the pores can influence the local equilibrium and kinetics of the transformation process. Studying the reduction of iron oxide by hydrogen helps us understand the sluggish reduction process and its significance for sustainable steelmaking.
PHYSICAL REVIEW LETTERS
(2023)
Article
Nanoscience & Nanotechnology
Se-Ho Kim, Kihyun Shin, Xuyang Zhou, Chanwon Jung, Hyun You Kim, Stella Pedrazzini, Michele Conroy, Graeme Henkelman, Baptiste Gault
Summary: Atom probe tomography is a useful technique for obtaining sub-nanoscale information from technologically-relevant materials. However, the analysis of functional ceramics, especially perovskites, remains challenging due to low yield and success rate. This study shows that a metallic coating can prevent charge penetration and suppress the volume change associated with the piezoelectric effect, allowing for successful analysis of BaTiO3 particles in a metallic matrix.
SCRIPTA MATERIALIA
(2023)
Article
Chemistry, Multidisciplinary
Yan Ma, Jae Wung Bae, Se-Ho Kim, Matic Jovicevic-Klug, Kejiang Li, Dirk Vogel, Dirk Ponge, Michael Rohwerder, Baptiste Gault, Dierk Raabe
Summary: The authors demonstrate a method of sustainable steel production by reducing solid iron oxides with hydrogen released from ammonia. Ammonia, which can be synthesized with green hydrogen and release hydrogen again through the reduction reaction, connects with green iron making as a replacement for fossil reductants. The study shows that ammonia-based reduction of iron oxide is as effective as hydrogen-based direct reduction, and can be implemented industrially with existing technologies.
Article
Chemistry, Physical
Leonardo Shoji Aota, Chanwon Jung, Siyuan Zhang, Se-Ho Kim, Baptiste Gault
Summary: Pd-based electro-catalysts play a crucial role in enhancing the methanol oxidation reaction kinetics in alcohol fuel cells, but their performance degrades over time. Scanning electron microscopy and atom probe tomography were used to investigate the chemical changes at the microstructural/atomic scale responsible for this effect after accelerated degradation tests. No morphological changes were observed after 1000 MOR cycles, but leaching of Pd and B from PdAu nanoparticles and the formation of Au-rich regions on the catalyst's surface were identified. These insights underscore the importance of understanding the chemical modifications during MOR for the design of new catalysts.
ACS ENERGY LETTERS
(2023)
Article
Chemistry, Physical
Su-Hyun Yoo, Leonardo Shoji Aota, Sangyong Shin, Ayman A. El-Zoka, Phil Woong Kang, Yonghyuk Lee, Hyunjoo Lee, Se-Ho Kim, Baptiste Gault
Summary: The introduction of interstitial dopants offers a new approach to optimize the catalytic activity of nanoparticles. However, the stability of a property-enhancing dopant (B) introduced in the controlled synthesis of a Pd aerogel electrocatalyst is significantly reduced after the hydrogen oxidation reaction. First-principles calculations suggest that the presence of H on the surface leads to the departure of subsurface B from the Pd nanostructure. This destabilization of subsurface B is more pronounced with increased H occupation of surface and interstitial sites. Thus, H2 fuel itself promotes the microstructural degradation and activity drop of the electrocatalyst.
ACS ENERGY LETTERS
(2023)
Article
Green & Sustainable Science & Technology
Mahander P. P. Singh, Se-Ho Kim, Xuyang Zhou, Hiram Kwak, Alisson Kwiatkowski da Silva, Stoichko Antonov, Leonardo Shoji Aota, Chanwon Jung, Yoon Seok Jung, Baptiste Gault
Summary: This study uses cryogenic atom probe tomography to investigate the surface chemistry changes of cathode materials for Li-ion batteries. The formation of Li2CO3 species on the surface of a LiNi0.8Mn0.1Co0.1O2 (NMC811) cathode material when exposed to air is observed. These findings are crucial for improving cathode synthesis and cell assembly protocols, as well as understanding cathode degradation processes.
ADVANCED ENERGY AND SUSTAINABILITY RESEARCH
(2023)
Article
Chemistry, Physical
Se-Ho Kim, Kang Dong, Huan Zhao, Ayman A. El-Zoka, Xuyang Zhou, Eric V. Woods, Finn Giuliani, Ingo Manke, Dierk Raabe, Baptiste Gault
Summary: By utilizing cryo-atom probe tomography, we have identified the degradation mechanisms of the liquid electrolyte, Si electrode, and their interface. The corrosion of the Si anode results from the decomposition of Li salt even before charge-discharge cycles. Volume shrinkage during delithiation leads to the formation of nanograins through recrystallization. These newly formed grain boundaries facilitate the pulverization of nanoscale Si fragments, some of which float in the electrolyte. Phosphorus is segregated to these grain boundaries, confirming the decomposition of the electrolyte. These findings contribute to an understanding of the self-catalyzed/accelerated degradation of Si anodes and can inform the development of new battery designs that are unaffected by these limiting factors.
JOURNAL OF PHYSICAL CHEMISTRY LETTERS
(2022)
Article
Materials Science, Multidisciplinary
Baptiste Gault, Kevin Schweinar, Siyuan Zhang, Leopold Lahn, Christina Scheu, Se-Ho Kim, Olga Kasian
Summary: The search for a new energy paradigm with net-zero carbon emissions requires cross-disciplinary technologies in engineering, chemistry, physics, surface, and materials sciences. To improve device performance and lifetime, establishing a scientific foundation to guide material design and bridging microscopy and spectroscopy techniques is necessary.