Skeleton-Based Conditionally Independent Gaussian Process Implicit Surfaces for Fusion in Sparse to Dense 3D Reconstruction

Abstract

3D object reconstructions obtained from 2D or 3D cameras are typically noisy. Probabilistic algorithms are suitable for information fusion and can deal with noise robustly. Consequently, these algorithms can be useful for accurate surface reconstruction. This paper presents an approach to estimate a probabilistic representation of the implicit surface of 3D objects. One of the contributions of the paper is the pipeline for generating an accurate reconstruction, given a set of sparse points that are close to the surface and a dense noisy point cloud. A novel submapping method following the topology of the object is proposed to generate conditional independent Gaussian Process Implicit Surfaces. This allows inference and fusion mechanisms to be performed in parallel to later propagating information through the submaps. Large datasets can efficiently be processed by the proposed pipeline producing not only a surface but also the uncertainty information of the reconstruction. We evaluate the performance of our algorithm using simulated and real datasets.

Online Informative Path Planning for Active Information Gathering of a 3D Surface

“…However, the normal GP formulation is limited to 2D terrains or 3D Euclidean space. To facilitate mapping 3D surfaces, Gaussian process implicit surfaces (GPISs) [23] have been used for surface reconstruction [24], [25], [26], object shape estimation [27], and pipeline thickness mapping [28]. The key idea is to represent the surface using a function that specifies whether a point in space is on the surface, outside the surface, or inside the surface.…”

Gaussian Process Gradient Maps for Loop-Closure Detection in Unstructured Planetary Environments

“…GP elevation maps in [24] are used as a prior for Bayesian fusion. Other GP-based mapping properties such as occupancy [25], thickness [26], implicit surface [27], [28], among others, have been extensively studied in recent years. Most of these works exploit the probabilistic nature and the inference capabilities to fill-up areas of missing data, to produce maps of a desirable resolution, to filter noise or for data fusion.…”

Active and Interactive Mapping with Dynamic Gaussian Process Implicit Surfaces for Mobile Manipulators

“…The Gaussian Probabilistic Implicit Surface (GPIS) [15], on the other hand, offers a probabilistic yet accurate map representation of the scene. The GPIS encodes spatial correlation embedded in input data and it is suitable for Bayesian fusion on scans from varying viewpoints [16], [17]. One can query the GPIS for scene geometry and uncertainty at arbitrary resolution, visited or unexplored alike.…”

“…Being a dense implicit surface, GPIS facilitate obstacle avoidance for safe pick trajectory planning, it also offers a natural mechanism to check for object manipulability in space. The surface normals, available as a natural component of GPIS with derivatives [16], [17] can be exploited for interactive applications [18], including object shape detection, classification and validation. On the other hand, the probabilistic formulation in GPIS makes it particularly amenable to active perception problems as the robot can use any IG based analytical methods to make flexible decision about the next optimal move.…”

“…Initially, all cells are assigned with negative infinity time corresponding to no penalty. As a failure occurs, the t f [s i ] at the robot's position is set to the current time, (16) to ensure the cell is excluded from NBV selection for a small time window.…”