Voting-Based Pose Estimation Using 3D Sensor Pair Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Pose estimation from 3D sensor data is challenging due to sensor noise, occlusions, and clutter, especially for industrial parts with planar surfaces, where existing methods using oriented surface points are not discriminative enough, and require dense point clouds.
Innovation Solution
A voting-based method using pair features that include oriented surface points, oriented boundary points, and boundary line segments, which are more compact and informative, enabling accurate and efficient pose estimation by leveraging geometric primitives such as surface pair features, boundary pair features, and line pair features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If oriented surface points are used for pose estimation, then the method works for objects with significant curvature changes, but the features are not compact and discriminative enough for planar industrial objects
Solution Approach 1:
The patent segments the object surface into distinct feature types: oriented surface points for curved regions and oriented boundary points for planar regions. This segmentation allows each feature type to be optimized for its specific geometric context, resolving the contradiction between versatility for curved objects and discriminative power for planar objects.
Solution Approach 2:
The patent applies different feature extraction strategies to different local regions of the object surface. Oriented boundary points with direction vectors are used specifically for planar surfaces where they provide compact representation, while oriented surface points are used for curved surfaces. This local quality approach ensures optimal performance for each object type.
2Measurement precision
If dense point clouds are used for pose estimation, then feature matching accuracy improves, but computation time increases and the method becomes sensitive to noise and occlusions
Solution Approach 1:
The patent extracts only the essential boundary information from the 3D point cloud by identifying and representing object boundaries as oriented boundary points and line segments. This extraction process removes redundant interior points while preserving the discriminative geometric features needed for accurate pose estimation, thereby reducing computation time without sacrificing matching accuracy.
Solution Approach 2:
The patent transitions from representing objects using dense 3D point clouds to representing them using 1D boundary curves and 2D surface patches with oriented normals. This dimensional reduction transforms the problem from matching numerous 3D points to matching compact boundary features, significantly reducing computational complexity while maintaining or improving discriminative power.
3Adaptability or versatility
If 3D sensor data is used without images or prior information, then the method works for unknown objects, but points, lines, and planes are not very discriminative individually and require combinatorial matching
Solution Approach 1:
The patent merges multiple geometric primitives (oriented boundary points, line segments, and surface patches) into a unified feature representation system. By combining these different feature types with their associated orientation and geometric constraints, the method creates highly discriminative composite features that can be matched individually without requiring complex combinatorial matching, thus reducing overall system complexity.
Data Source
AI summary
A pose of an object is estimated by first defining a set of pair features as pairs of geometric primitives, wherein the geometric primitives include oriented surface points, oriented boundary points, and boundary line segments. Model pair features are determined based on the set of pair features for a model of the object. Scene pair features are determined based on the set of pair features from data acquired by a 3D sensor, and then the model pair features are matched with the scene pair features to estimate the pose of the object.