Surface Reconstruction Using Pose Graph Anchored Volumes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing surface reconstruction methods face challenges in large-scale environments due to cumulative tracking and mapping errors, leading to inconsistent, gap-filled, or overlapping data, which results in resource-intensive and artifact-prone reconstructions.

Innovation Solution

The method involves creating multiple separate volumes for surface reconstruction, anchoring each vertex to nodes or combinations of nodes in a pose graph, allowing volumes to deform and share vertices to avoid gaps and overlaps, and updating the pose graph to maintain accurate surface reconstruction across volumes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Area of stationary object

If multiple depth maps are obtained and additional volumes are defined to reconstruct large-scale environments, then the coverage area increases, but cumulative errors in volume positions accumulate leading to inconsistent surface reconstruction

Engineering Contradiction:
Improvecoverage areaVSAvoidsurface reconstruction accuracy
Core Design Contradiction:
Area of stationary objectVSManufacturing precision

Solution Approach 1:

The large-scale environment is divided into multiple overlapping local volumes, each reconstructed independently with high precision. These segmented volumes are then integrated through pose graph optimization, allowing each segment to maintain its local accuracy while contributing to the overall large-scale coverage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

A pose graph serves as an intermediary structure that connects multiple local volumes through shared vertices and correspondences. The pose graph optimization adjusts the relative poses of volumes and re-anchors vertices to minimize cumulative errors, enabling accurate integration of segmented reconstructions across large areas.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Manufacturing precision

If volumes are kept separate to maintain individual accuracy, then local reconstruction precision is improved, but gaps and overlaps occur at volume boundaries creating visible artifacts

Engineering Contradiction:
Improvelocal reconstruction precisionVSAvoidgaps and overlaps at boundaries
Core Design Contradiction:
Manufacturing precisionVSObject-generated harmful factors

Solution Approach 1:

Adjacent volumes are merged through shared vertices that are anchored to common pose graph nodes. This merging ensures continuity at boundaries by enforcing vertex correspondence between neighboring volumes, eliminating gaps and overlaps while preserving the local precision achieved through independent volume reconstruction.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The volume boundaries are made dynamic through the pose graph optimization process. As the pose graph is updated with new correspondences and optimized, the relative positions and orientations of volumes adjust dynamically, allowing vertices at boundaries to be re-anchored to new nodes that minimize errors and maintain smooth transitions between adjacent volumes.

Inventive Principle:
Principle #15Dynamics

3Manufacturing precision

If pose graph is frequently updated to reduce cumulative errors, then surface reconstruction accuracy is improved, but computational resources and processing time increase

Engineering Contradiction:
Improvesurface reconstruction accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Manufacturing precisionVSUse of energy by moving object

Solution Approach 1:

The pose graph is updated periodically rather than continuously with every new depth map. This periodic updating reduces computational overhead by performing full optimization at intervals while maintaining accurate surface reconstruction through incremental updates to the pose graph structure between full optimizations.

Inventive Principle:
Principle #19Periodic action

Solution Approach 2:

Vertices are pre-anchored to pose graph nodes during volume creation, establishing a stable reference framework before optimization is needed. This preliminary anchoring reduces the computational complexity of subsequent pose graph updates, as the vertex-node correspondences are already established and only the pose parameters need adjustment.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3195270B1Using free-form deformations in surface reconstruction
Publication Date: 2018.10.24 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP3195270B1 patent drawingFigure 1
  • EP3195270B1 patent drawingFigure 2
  • EP3195270B1 patent drawingFigure 3

AI summary

Volumes of a 3D physical space are used in a surface reconstruction process, where adjacent volumes share vertices so that no gaps or overlaps between the volumes exist. As a result, a continuous surface is obtained in the surface reconstruction process. The vertices are anchored to nodes in a pose graph, such that locations of the vertices are adjusted as the pose graph is updated. As a result, a deformation of the volumes is permitted. Based on the deformation of a volume, a region of a depth map of the physical space is deformed correspondingly. Each vertex can be anchored to a closest node of the pose graph, or to a point which is based on a combination of nodes. In one approach, the point is defined based on the closest node and other nodes within a defined radius of the closest node.