Point Cloud Annotation via Likelihood-Based Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual annotation of three-dimensional point clouds is time-consuming and labor-intensive, especially in complex scenes with overlapping structures, due to the need for precise selection and division of viewpoint and gazing points, and existing methods either require extensive click operations or suffer from poor accuracy in area division.
Innovation Solution
A point cloud annotation device and method that calculates an evaluation function to cluster three-dimensional points, allowing for batch annotation by displaying clusters in descending order of likelihood, enabling easy selection and annotation of target objects with reduced operational burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation is performed by visually selecting point clouds in CAD software, then annotation can be performed on three-dimensional point clouds, but the operating time becomes long due to the need to repeatedly find and cut out directions where subjects do not overlap
Solution Approach 1:
The point cloud is automatically divided into multiple divided areas, allowing the annotation system to process and display multiple regions simultaneously. This segmentation enables annotators to work on multiple non-overlapping subjects in parallel views, eliminating the need to repeatedly switch viewpoints and manually search for suitable annotation directions.
Solution Approach 2:
The system transitions from single-viewpoint three-dimensional visualization to multi-viewpoint simultaneous display, adding the dimension of multiple observation angles. By displaying multiple divided areas with different viewpoint directions at once on the screen, the system allows annotators to select and annotate subjects without repeatedly changing viewpoints, significantly reducing operating time while maintaining annotation accuracy.
2Productivity
If the number of divided areas is increased to cover more regions, then more areas can be annotated at once, but the number of click operations increases and working time increases
Solution Approach 1:
Multiple divided areas are merged into a single unified display interface, allowing annotators to select and annotate multiple regions through a streamlined process. The system combines the functionality of viewing and annotating multiple areas into one integrated workflow, reducing the number of separate click operations needed while maintaining high annotation throughput.
3Ease of manufacture
If automatic division of point clouds is performed, then the point cloud can be divided into multiple areas, but when subjects are dense or in contact with surrounding structures, the number of divided areas increases and requires more manual correction operations
Solution Approach 1:
The system provides dynamic control over the division process, allowing annotators to adjust viewpoint directions and gazing points interactively. This dynamic adjustment capability enables optimization of divided areas to reduce unnecessary fragmentation in dense scenes, minimizing the need for manual correction operations while maintaining the benefits of automatic division.
Data Source
AI summary
Annotation can be easily performed on a three-dimensional point cloud and a working time can be reduced. An interface unit 22 displays a point cloud indicating a three-dimensional point on an object, and receives designation of a three-dimensional point indicating an annotation target object and designation of a three-dimensional point not indicating the annotation target object. A candidate cluster calculation unit 32 calculates a value of a predetermined evaluation function indicating a likelihood of a point cloud cluster being the annotation target object based on the designation of a three-dimensional point for point cloud clusters obtained by clustering the point clouds. A cluster selection and storage designation unit 34 causes the interface unit 22 to display the point cloud clusters in descending order of the value of the evaluation function, and receives a selection of a point cloud cluster to be annotated. An annotation execution unit 36 executes annotation indicating the annotation target object for each three-dimensional point included in the selected point cloud cluster.