3D Point Cloud Labeling via Relative Coordinates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for semantic segmentation of three-dimensional point groups face challenges in accurately identifying class labels, especially for high-density point groups with complex shapes and near object boundaries, due to pre-processing divisions and reduced sample sizes, leading to erroneous identifications.
Innovation Solution
A learning device that acquires and processes data for sampled three-dimensional points, using a class label assigning model comprising multiple neural networks to convert relative coordinates and feature quantities, enabling accurate assignment of class labels by considering neighboring points' validity and spatial distribution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If pre-processing division and sampling are performed on high-density three-dimensional point groups, then processing time and computational load are reduced, but measurement precision and identification accuracy deteriorate due to loss of spatial distribution information
Solution Approach 1:
The patent creates a virtual three-dimensional point group that copies and represents the spatial distribution characteristics of the original high-density point group. By using virtual points with coordinates calculated based on relative position relationships, the system preserves the essential spatial information needed for accurate identification while working with a smaller, more manageable dataset.
Solution Approach 2:
The patent introduces a conversion model as an intermediary that transforms relative coordinates from the original point group into a virtual point group structure. This conversion model acts as a mediator that maintains the spatial distribution information while enabling efficient processing through reduced sample sizes.
2Device complexity
If the number of samples is reduced through sampling, then device complexity and processing requirements are decreased, but reliability deteriorates due to insufficient representation of complex shapes and boundaries
Solution Approach 1:
The patent changes the parameter representation from absolute coordinates to relative coordinates. By expressing point positions as relative coordinates with respect to reference points, the system preserves the geometric relationships and spatial distribution patterns essential for identifying complex shapes and boundaries, even with reduced sample sizes.
Solution Approach 2:
The patent adds a relational dimension to the data structure by introducing relative coordinate systems. Instead of treating points in isolation, the system considers their relationships to reference points, effectively adding a dimensional layer of information about spatial relationships that compensates for reduced sample counts.
3Productivity
If conventional convolution methods are used on sampled points, then processing speed is improved, but manufacturing precision deteriorates due to erroneous class label assignments near object boundaries
Solution Approach 1:
The patent applies different processing approaches to different regions of the point group. By identifying reference points and calculating relative coordinates locally with respect to these references, the system adapts the processing method to preserve local spatial relationships, which is particularly important for accurately identifying points near object boundaries.
Solution Approach 2:
The patent performs preliminary conversion of coordinates and selection of reference points before applying convolution operations. By pre-organizing the data structure with reference points and relative coordinates, the system prepares the data in a way that maintains spatial relationships, enabling more accurate class label assignments during subsequent processing steps.
Data Source
AI summary
An identification device acquires a plurality of identification target points by sampling a target point group that is a set of three-dimensional target points. The identification device calculates relative coordinates of a neighboring point of the identification target point with respect to the identification target point. The identification device inputs coordinates of the plurality of identification target points and relative coordinates of neighboring points with respect to each of the plurality of identification target points into a class label assigning learned model to acquire class labels of the plurality of identification target points and validity of the class labels with respect to the neighboring points for each of the plurality of identification target points. The identification device assigns the class labels to the plurality of identification target points, assigns the class labels to the neighboring points for each of the plurality of identification target points when the validity of the class label is included in a range determined by a predetermined threshold value, and identifies the class labels of the identification target point and the neighboring point.