Article
Computer Science, Information Systems
Munkhtulga Byambaa, Gou Koutaki, Lodoiravsal Choimaa
Summary: This study proposes a method for 6D pose estimation of transparent objects, which uses a deep neural network to estimate 2D keypoints and uses the PnP algorithm to estimate the 6D pose of the object. The experiments demonstrate that the method is capable of grasping transparent objects from different backgrounds and outperforms other 6D pose estimation methods, with the potential to be applied to real-world images.
Article
Automation & Control Systems
Yaoxian Song, Jun Wen, Dongfang Liu, Changbin Yu
Summary: Vision-based robotic grasping is a fundamental task in robotic control. This paper proposes a novel multi-modal neural network to predict grasps in real-time. By hierarchically fusing RGB and depth information and quantifying the uncertainty of depth data, the proposed method improves grasping performance and reduces the influence of incomplete and low-quality raw data.
INTERNATIONAL JOURNAL OF CONTROL AUTOMATION AND SYSTEMS
(2022)
Review
Computer Science, Artificial Intelligence
Guoguang Du, Kai Wang, Shiguo Lian, Kaiyong Zhao
Summary: This paper provides a comprehensive overview of vision-based robotic grasping, focusing on key tasks such as object localization, object pose estimation, and grasp estimation. Various methods combining these tasks, including traditional and deep learning-based approaches, are reviewed in detail. Challenges and future directions in the field are also highlighted.
ARTIFICIAL INTELLIGENCE REVIEW
(2021)
Article
Automation & Control Systems
Xuebing Liu, Xiaofang Yuan, Qing Zhu, Yaonan Wang, Mingtao Feng, Jiaming Zhou, Zhen Zhou
Summary: This paper proposes a depth adaptive feature extraction and dense prediction network for estimating the pose of an object in industrial robotic grasping. The network utilizes the two-modal data, color image and depth image, to fuse multimodal textures and retain the 3D structure. A dense prediction strategy is adopted to regress the object pose, mitigating the instability caused by outliers.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Automation & Control Systems
Hui Zhang, Zhicong Liang, Chen Li, Hang Zhong, Li Liu, Chenyang Zhao, Yaonan Wang, Q. M. Jonathan Wu
Summary: Pose estimation is a critical technology in industrial robotics. This study proposes a practical robotic grasping method that uses 6D pose estimation with protective correction to address the challenges of rapid detection in complex multiscene environments. The method trains a deep object pose estimation network with a synthetic dataset and uses the perspective-n-point algorithm to estimate the 6-DoF pose. A corrected grasping pose algorithm is also proposed to prevent collisions caused by misrecognition.
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS
(2022)
Article
Robotics
Xiaozheng Liu, Yunzhou Zhang, Dexing Shan
Summary: Semantic segmentation is important for robots to perceive and interact with the environment. However, accurate segmentation is challenging in the presence of unseen objects. To address this, we propose a few-shot semantic segmentation framework for robot perception, integrating segmentation and grasp pose detection. Our method quickly identifies unseen targets and achieves accurate results with only a few labeled support images. Experimental results and real robotic grasping experiments demonstrate the effectiveness of our approach.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Plant Sciences
Da-Young Lee, Dong-Yeop Na, Carlos Gongora-Canul, Sriram Baireddy, Brenden Lane, Andres P. Cruz, Mariela Fernandez-Campos, Nathan M. Kleczewski, Darcy E. P. Telenko, Stephen B. Goodwin, Edward J. Delp, C. D. Cruz
Summary: Quantification of corn tar spot symptoms was traditionally done through visual-based estimations of leaf area covered by pathogenic structures. This study introduced a contour-based detection method using RGB images to quantify disease intensity, showing potential utility in complementing traditional visual-based severity estimations and building an accurate high-throughput pipeline for tar spot symptom scoring.
FRONTIERS IN PLANT SCIENCE
(2021)
Article
Computer Science, Interdisciplinary Applications
Chungang Zhuang, Zhe Wang, Heng Zhao, Han Ding
Summary: This article proposes a Semantic Point Pair Feature (PPF) method for 3D object pose estimation, which combines semantic image segmentation using deep learning with voting-based 3D object pose estimation. The method improves the robustness and efficiency of 3D object pose estimation in cluttered scenes with partial occlusions.
ROBOTICS AND COMPUTER-INTEGRATED MANUFACTURING
(2021)
Article
Engineering, Electrical & Electronic
Huei-Yung Lin, Shih-Cheng Liang, Yu-Kai Chen
Summary: This paper introduces a robotic grasping system with multi-view depth image acquisition, utilizing a series of algorithms for noise removal and target pose estimation to ultimately increase grasping efficiency and demonstrate feasibility through experimentation.
IEEE SENSORS JOURNAL
(2021)
Article
Engineering, Electrical & Electronic
Yaowei Li, Fei Guo, Miaotian Zhang, Shuangfu Suo, Qi An, Jinlin Li, Yang Wang
Summary: In this study, a vision-based intelligent robotic grasping system was developed using deep learning to obtain the 6D pose of an axisymmetric body in industrial stacked scenarios. A multitask real-time convolutional neural network, Key-Yolact, was proposed to detect keypoints, perform object detection, instance segmentation, and multiobject 2D keypoint detection. The system showed practical tradeoff between inference speed and precision, making it effective for industrial scenarios.
Article
Computer Science, Artificial Intelligence
Shuangjun Liu, Xiaofei Huang, Nihang Fu, Cheng Li, Zhongnan Su, Sarah Ostadabbas
Summary: The computer vision field has made great progress in interpreting semantic meanings from images, but its algorithms can be fragile for tasks that have adverse vision conditions and limited data/label pairs. One such task is in-bed human pose monitoring, which has significant value in healthcare applications. However, the lack of publicly available in-bed pose datasets hinders the applicability of successful human pose estimation algorithms. In this paper, we introduce the Simultaneously-collected multimodal Lying Pose (SLP) dataset, which includes in-bed pose images captured using multiple imaging modalities and propose a physical hyper parameter tuning strategy for generating ground truth pose labels under adverse vision conditions. The SLP dataset is compatible with mainstream human pose datasets, allowing the effective training of state-of-the-art 2D pose estimation models with promising performance.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Software Engineering
Xinyue Zhao, Quanzhi Li, Yue Chao, Quanyou Wang, Zaixing He, Dong Liang
Summary: This article introduces a new dataset, RT-Less, for pose estimation research on reflective texture-less objects. The dataset contains 38 reflective metal parts, providing a large number of real and synthetic images, as well as accurate ground truth poses, bounding-box annotations, and masks. Baseline results and an iterative pose optimization method are also provided.
Article
Chemistry, Analytical
Hao Xu, Qi Sun, Weiwei Liu, Minghao Yang
Summary: This paper proposes a multi-task secure grasping detection model, consisting of a grasping relationship network and an oriented rectangles detection network. Experiments demonstrate that the proposed method outperforms existing methods in detecting grasp positions and stacking relationships of objects in complex scenes, showing good applicability on robot platforms.
Article
Computer Science, Artificial Intelligence
Yu-Huan Wu, Yun Liu, Jun Xu, Jia-Wang Bian, Yu-Chao Gu, Ming-Ming Cheng
Summary: This article introduces a novel network, MobileSal, for efficient RGB-D SOD, achieving good results in terms of computational efficiency and feature representation capability.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Jinming Cao, Hanchao Leng, Daniel Cohen-Or, Dani Lischinski, Ying Chen, Changhe Tu, Yangyan Li
Summary: A novel method is proposed in this paper to fuse RGB and depth information, simplifying the bridge between RGB and RGBD semantic segmentation and avoiding the need for more complex network structures. Experimental results show that the proposed method consistently outperforms baseline models.