TEASER: Fast and Certifiable Point Cloud Registration
نویسندگان
چکیده
We propose the first fast and certifiable algorithm for registration of two sets three-dimensional (3-D) points in presence large amounts outlier correspondences. A certifiable algorithm is one that attempts to solve an intractable optimization problem (e.g., robust estimation with outliers) provides readily checkable conditions verify if returned solution optimal produced most accurate estimate face or bound its suboptimality accuracy. Toward this goal, we reformulate using a xmlns:xlink="http://www.w3.org/1999/xlink">truncated least squares (TLS) cost makes insensitive fraction spurious Then, provide general graph-theoretic framework decouple scale, rotation, translation estimation, which allows solving cascade three transformations. Despite fact each subproblem (scale, estimation) still nonconvex combinatorial nature, show 1) TLS scale (component-wise) can be solved polynomial time via xmlns:xlink="http://www.w3.org/1999/xlink">adaptive voting scheme, 2) rotation relaxed semidefinite program (SDP) relaxation tight, even extreme rates, 3) drastic pruning outliers by finding maximum clique. name resulting TEASER ( xmlns:xlink="http://www.w3.org/1999/xlink">Truncated squares Estimation And SEmidefinite Relaxation ). While SDP relaxations typically slow, develop second algorithm, named TEASER++, uses xmlns:xlink="http://www.w3.org/1999/xlink">graduated nonconvexity leverages xmlns:xlink="http://www.w3.org/1999/xlink">Douglas-Rachford Splitting efficiently certify global optimality. For both algorithms, theoretical bounds on errors, are their kind problems. Moreover, test performance standard benchmarks, object detection datasets, xmlns:xlink="http://www.w3.org/1999/xlink"/> 3DMatch scan matching dataset, algorithms dominate state-of-the-art RANSAC, branch-&-bound, heuristics) more than $\text{99}\%$ when known, TEASER++ run milliseconds it currently fastest so also problems without correspondences hypothesizing all-to-all correspondences), where largely outperforms ICP Go-ICP while being orders magnitude faster. release open-source C++ implementation TEASER++.
منابع مشابه
Colored Point Cloud Registration Revisited Supplementary Material
As in Section 4.3, this objective is minimized using the Gauss-Newton method. Specifically, we start from an initial transformation T and perform optimization iteratively. In each iteration, we locally parameterize T with a 6-vector ξ, evaluate the residual r and Jacobian Jr at T, solve the linear system in (21) to compute ξ, and use ξ to update T. To compute the Jacobian, we need the partial d...
متن کاملCloud To Cloud Registration For 3d Point Data
Grant, Darion Shawn. Ph.D., Purdue University, December 2013. Cloud To Cloud Registration For 3D Point Data. Major Professors: James Bethel and Melba Crawford. The vast potential of digital representation of objects by large collections of 3D points is being recognized on a global scale and has given rise to the popularity of point cloud data (PCD). 3D imaging sensors provide a means for quickl...
متن کاملHierarchical Registration Method for Airborne and Vehicle LiDAR Point Cloud
A new hierarchical method for the automatic registration of airborne and vehicle light detection and ranging (LiDAR) data is proposed, using three-dimensional (3D) road networks and 3D building contours. Firstly, 3D road networks are extracted from airborne LiDAR data and then registered with vehicle trajectory lines. During the registration of airborne road networks and vehicle trajectory line...
متن کاملRobust Automatic 3D Point Cloud Registration and Object Detection
In order to construct survey-grade 3D models of buildings, roads, railway stations, canals and other similar structures, the 3D environment must be fully recorded with accuracy. Following this, accurate measurements of the dimensions can be made on the recorded 3D datasets to enable 3D model extraction without having to return to the site and in significantly reduced times. The model may be com...
متن کاملAlmost-Isometries for Point Cloud and Image Registration
Matching sets of points in 3D point clouds and aligning images are problems that are common to robotics, GIS, and medical imaging. The transformations resulting from the alignments can aid in areas ranging from robot localization to the measuring of what has changed between images, such as landscape developments or the healing after a surgery. The main contribution of this work is the implement...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Transactions on Robotics
سال: 2021
ISSN: ['1552-3098', '1941-0468', '1546-1904']
DOI: https://doi.org/10.1109/tro.2020.3033695