3D Point Cloud Feature Detection for Robotic Guidance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current feature detection systems in manufacturing environments face challenges in accurately identifying predefined geometric shapes in noisy 3D point clouds, which hinders precise robotic operations, especially in areas where human intervention is hazardous or impractical.
Innovation Solution
A feature detection system utilizing two electromagnetic sensors, such as visible light cameras, to generate a 3D point cloud by creating a disparity map and performing plane fitting to identify predefined geometric shapes before projecting boundary pixels onto these shapes, thereby reducing noise and enhancing accuracy in feature localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature detection methods are used on noisy 3D point clouds, then feature detection can be performed, but the accuracy of identifying predefined geometric shapes deteriorates
Solution Approach 1:
The patent applies preliminary action by performing plane fitting to identify predefined geometric shapes in the 3D point cloud before conducting edge detection. This preliminary identification of geometric structures (such as planes representing component surfaces) establishes a clean reference framework that filters out noise, allowing subsequent boundary pixel projection to accurately locate features relative to these identified shapes rather than working directly with the noisy point cloud data.
2Manufacturing precision
If geometric shapes are identified directly in noisy 3D point clouds, then feature detection is possible, but the precision of shape identification deteriorates
Solution Approach 1:
The patent uses identified geometric shapes as an intermediary between the noisy 3D point cloud and the final feature detection. The plane fitting process creates idealized geometric models that serve as mediators, filtering out noise while preserving essential structural information. Boundary pixels are then projected onto these intermediary geometric shapes, allowing precise feature localization without direct contamination from the noisy point cloud data.
3Measurement precision
If boundary pixels are projected directly onto 3D point clouds, then feature locations can be identified, but the accuracy of localization deteriorates due to noise
Solution Approach 1:
The patent applies preliminary action by first identifying geometric shapes through plane fitting before projecting boundary pixels. This preliminary geometric modeling creates a noise-filtered reference framework, so when boundary pixels are projected onto the identified geometric shapes rather than directly onto the noisy point cloud, the feature localization accuracy is significantly improved while eliminating noise interference.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Aspects herein use a feature detection system to visually identify a feature on a component. The feature detection system includes at least two cameras that capture images of the feature from different angles or perspectives. From these images, the system generates a 3D point cloud of the components in the images. Instead of projecting the boundaries of features onto the point cloud directly, the aspects herein identify predefined geometric shapes in the 3D point cloud. The system then projects pixel locations of the feature's boundaries onto the identified geometric shapes in the point cloud. Doing so yields the 3D coordinates of the feature which then can be used by a robot to perform a manufacturing process.