Deep Learning and OcTree-GPU-Based ICP for Efficient 6D Model Registration of Large Objects

Ahmadli, Ismayil; Bedada, Wendwosen Bellete; Palli, Gianluca

doi:10.1007/978-3-030-96359-0_3

Cited by 3 publications

(2 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, deep learning frameworks have been integrated with fine registration methods such as ICP and NDT to provide faster convergence without the requirement of a good initial pose. Ahmdali et al [15] use deep learning to match features in the source and reference to decrease computation time and provide a coarse guess for the finer registration process using ICP. The solution provides improved performance with significant occlusions and minimal overlap.…”

Section: Learning-based Approachesmentioning

confidence: 99%

Novel Approaches for Point Cloud Analysis with Evidential Methods: A Multifaceted Approach to Object Pose Estimation, Point Cloud Odometry, and Sensor Registration

Bhandari,

Phillips,

McAree

2024

Point Cloud Generation and Its Applications [Working Title]

View full text Add to dashboard Cite

Autonomous agents must understand their environment to make decisions. Perception systems often interpret point cloud measurements to extract beliefs about their surroundings. A common strategy is to seek beliefs that are least likely to be false, commonly known as cost-based approaches. These metrics have limitations in practical applications, such as in the presence of noisy measurements, dynamic objects, and debris. Modern solutions integrate additional stages such as segmentation to counteract these limitations, thereby increasing the complexity of the algorithms while being internally flawed. An alternative strategy is to extract beliefs that are best supported by the data. We call these evidence-based methods. This difference allows for robustness to the limitations of using cost-based methods without needing complex additional stages. Essential perception tasks such as object pose estimation, point cloud odometry, and sensor registration are solved using evidence-based methods. The demonstrated approaches are simple, require minimum configuration and tuning, and circumvents the need for additional processing stages.

show abstract

Section: Learning-based Approachesmentioning

confidence: 99%

Novel Approaches for Point Cloud Analysis with Evidential Methods: A Multifaceted Approach to Object Pose Estimation, Point Cloud Odometry, and Sensor Registration

Bhandari,

Phillips,

McAree

2024

Point Cloud Generation and Its Applications [Working Title]

View full text Add to dashboard Cite

show abstract

“…Genetic algorithms have been employed to determine global extrema and provide accurate results by extending the vanilla ICP algorithm, as in [ 37 ]. Another subdivision of the research space has spawned with the recent developments in deep learning and the introduction of geometric deep learning [ 34 ], with recent effective registration extensions in [ 35 , 36 ]. New sensing technologies such as Aeva’s Doppler 4D LiDAR [ 38 ] have also prompted improvements in ICP for point cloud-to-point cloud registration with the Doppler Iterative Closest Point [ 39 ], which leverages additional metadata in the returned point cloud such as per point instantaneous velocities for accurate registration.…”

Section: Related Workmentioning

confidence: 99%

Real-Time 6-DOF Pose Estimation of Known Geometries in Point Cloud Data

Bhandari

Phillips

McĂree

2023

Sensors

View full text Add to dashboard Cite

The task of tracking the pose of an object with a known geometry from point cloud measurements arises in robot perception. It calls for a solution that is both accurate and robust, and can be computed at a rate that aligns with the needs of a control system that might make decisions based on it. The Iterative Closest Point (ICP) algorithm is widely used for this purpose, but it is susceptible to failure in practical scenarios. We present a robust and efficient solution for pose-from-point cloud estimation called the Pose Lookup Method (PLuM). PLuM is a probabilistic reward-based objective function that is resilient to measurement uncertainty and clutter. Efficiency is achieved through the use of lookup tables, which substitute complex geometric operations such as raycasting used in earlier solutions. Our results show millimetre accuracy and fast pose estimation in benchmark tests using triangulated geometry models, outperforming state-of-the-art ICP-based methods. These results are extended to field robotics applications, resulting in real-time haul truck pose estimation. By utilising point clouds from a LiDAR fixed to a rope shovel, the PLuM algorithm tracks a haul truck effectively throughout the excavation load cycle at a rate of 20 Hz, matching the sensor frame rate. PLuM is straightforward to implement and provides dependable and timely solutions in demanding environments.

show abstract

Development of a mobile-based intelligent module for identification and tracking of household appliances

Kaplun,

Shpakovskaya,

Sinitca

et al. 2023

Sci Rep

View full text Add to dashboard Cite

Every year manufacturers of household appliances improve their devices, trying to make everyday life easier for users. New smart devices have many useful features, but not all users can easily cope with the complexity of the devices. One of the main tasks of household appliance manufacturers is to ensure the convenience of using appliances, taking into account the increasing complexity. Therefore, any manufacturer supplies equipment with a short but useful instruction manual. Practice shows that no printed user manual can compare with a demonstration of the device operation by a professional consultant. Instructions for home appliances using augmented reality technology will allow users to get the necessary detailed information about the device in a short period of time. As part of this work, the task of developing an artificial intelligence-based module is being solved. This module consists of developed classification, matching, and tracking submodules that can provide simple and fast visual instructions to users of household appliances in real time. The identification of household appliances is performed with more than 0.9 accuracy, and the tracking inside an unidentified object using the camera of a mobile device is processed with the success score of about 0.68 and frames per second (FPS) about 7. Mobile applications based on the proposed intelligent modules for Android and iOS were developed.

show abstract

Deep Learning and OcTree-GPU-Based ICP for Efficient 6D Model Registration of Large Objects

Cited by 3 publications

References 18 publications

Novel Approaches for Point Cloud Analysis with Evidential Methods: A Multifaceted Approach to Object Pose Estimation, Point Cloud Odometry, and Sensor Registration

Novel Approaches for Point Cloud Analysis with Evidential Methods: A Multifaceted Approach to Object Pose Estimation, Point Cloud Odometry, and Sensor Registration

Real-Time 6-DOF Pose Estimation of Known Geometries in Point Cloud Data

Development of a mobile-based intelligent module for identification and tracking of household appliances

Contact Info

Product

Resources

About