REG-Net: Improving 6DoF Object Pose Estimation With 2D Keypoint Long-Short-Range-Aware Registration

Jiang, Junjie; He, Zaixing; Zhao, Xy; Zhang, Shuyou; Wu, Chenrui; Wang, Yan

doi:10.1109/tii.2022.3165979

Cited by 19 publications

(2 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the topic of RNNs in mobile robotics and laser scanner data manipulation, RNNs have been used to perform object recognition in point clouds [16], track objects in 2D point clouds [17], and more recently paired with a particle filter to monitor the localization of AMRs, as in [18]. To improve localization precision, [19] introduced a CNN for the 6DoF pose estimation of texture-less objects, and [20] proposed a deep supervised descent method with multiple seeds generation for 3D tracking in point clouds.…”

Section: Introductionmentioning

confidence: 99%

Cloud-Edge Architecture for the Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Mutti,

Pedrocchi,

Dimauro

2024

Preprint

View full text Add to dashboard Cite

The precise localization of mobile robots in unstructured environments is of utmost importance for many industrial and field applications, especially when the mobile robot is part of a more complex kinematic chain, such as a mobile manipulator. Being able to precisely localize affects the outcome of tasks that rely on an open-loop kinematic computation, such as work-station docking procedures. To achieve a repeatable and precise localization and positioning, mobile robots generally rely on onboard sensors, most commonly 2D laser scanners, whose readings are subjected to noise and numerous disturbing factors (e.g., materials reluctance). Problems arise when precise localization is needed in dynamic and unstructured environments where generally applicable methods won’t perform adequately or might be time-consuming to set up. In this work, we propose a cloud-edge computing architecture to deploy a recurrent neural network (RNN) based registration system, which uses a pair of consecutive LiDAR readings to estimate a fixed transformation. The capability of RNNs to process contiguous inputs will help neglect errors embedded in punctual laser scanner readings and output a more precise registration estimation. In such a way, the RNN can estimate a displacement error based on multiple consecutive readings and act as a sensor to be employed in a closed-loop control scheme. To tackle the dynamic and unstructured environments, the model is firstly tuned on synthetic LiDAR data to embed rigid transformations into the deep learning model, for then rapidly fine-tuned on local scenarios. After model architecture and optimization of hyperparameters, the devised model is tested in different scenarios, comparing the precise positioning capability of the AMR(autonomous mobile robot) with that of a classical registration algorithm. The results suggest that an RNN model can greatly improve the registration precision of laser scanner signals and, consequently, the precise positioning efficiency of AMRs.

show abstract

Section: Introductionmentioning

confidence: 99%

Cloud-Edge Architecture for the Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Mutti,

Pedrocchi,

Dimauro

2024

Preprint

View full text Add to dashboard Cite

show abstract

Section: Introductionmentioning

confidence: 99%

Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Mutti

Pedrocchi

Dimauro

2023

Preprint

View full text Add to dashboard Cite

The precise localization of mobile robots is of utmost importance for many industrial applications, especially when the mobile robot is part of a more complex kinematic chain, such as in a mobile manipulator. Furthermore, precise localization hugely affects the outcome of tasks that rely on an open-loop kinematic computation, such as work-station docking procedures. To achieve a repeatable and precise localization and positioning, mobile robots generally rely on onboard sensors, most commonly 2D laser scanners, whose readings are subjected to noise and numerous disturbing factors ( e.g., materials reluctance). In this work, we propose a recurrent neural network (RNN) based registration system, which uses a pair of consecutive LiDAR readings and estimates a fixed transformation. The capability of RNNs to process contiguous inputs will help neglect errors embedded in punctual laser scanner reading and output a more precise registration estimation. In such a way, the RNN can estimate a displacement error based on multiple consecutive readings and act as a sensor to be employed in a closed-loop control scheme. After a model architecture and optimization of hyperparameters, the devised model is tested in different scenarios, comparing the AMR precise positioning capability with a classical registration algorithm. The results suggest that an RNN model can greatly improve the registration precision of laser scanner signals and, consequently, the precise positioning efficiency of AMRs.

show abstract

Local geometric edge features based registration for textureless object in augmented reality assisted assembly

Li,

Wang,

Wei

2024

Virtual Reality

View full text Add to dashboard Cite

Image-based methods have been widely used in augmented reality (AR) assistant assembly systems. However, due to the lack of sufficient texture information on the surface of assembly part, traditional image feature matching methods still face challenges. This paper proposes a coarse-to-fine AR registration method for textureless assembly part. In the first stage, a new feature matching method which is called line neighborhood edge descriptor (LNED) is presented to find the coarse camera pose from textureless image. The LNED take the contour line of assembly part as the description object, and use local geometric edge of assembly part to describe the contour line. During the image matching, the binary encoding is used to reduce the computational consumption for LNED. In the second stage, spatial points in the CAD model of assembly part are reverse projected to the textureless image based on the coarse camera pose. And the bundle adjustment method based on the edge distance of the textureless image is adopted to iteratively calculate the precise camera pose. In the experimental evaluation, the proposed registration method shows high accuracy and fast speed in comparison with conventional registration methods, which demonstrates that our method can effectively solve the problem of AR registration for textureless assembly part.

show abstract

REG-Net: Improving 6DoF Object Pose Estimation With 2D Keypoint Long-Short-Range-Aware Registration

Cited by 19 publications

References 29 publications

Cloud-Edge Architecture for the Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Cloud-Edge Architecture for the Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Precise Positioning of Autonomous Mobile Robots through a Fine-Tuned RNN

Local geometric edge features based registration for textureless object in augmented reality assisted assembly

Contact Info

Product

Resources

About