Neural Network Building Scene Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer-aided design and engineering systems lack effective solutions for accurately segmenting building scenes, particularly in factory environments, where precise identification of partitions and junctions is crucial for tasks like automatic plant twin generation and production KPI monitoring.
Innovation Solution
A computer-implemented method using a neural network trained on top-down depth maps with labeled line segments and junctions to output a wireframe of building scenes, capable of detecting partitions and junctions, even in the presence of noise and distractors, allowing for automatic segmentation of building scenes from 3D point clouds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional computer-aided design and engineering systems are used for building scene segmentation, then the systems have established workflows and interfaces, but they lack accurate identification of partitions and junctions in building scenes
Solution Approach 1:
The patent introduces an intermediary processing pipeline between traditional CAD/CAE systems and building scene analysis. This pipeline includes depth map generation from point clouds, wireframe extraction algorithms, and automated partition/junction identification that bridges the gap between existing design systems and building scene segmentation requirements, enabling accurate identification without replacing core system workflows
Solution Approach 2:
The patent replaces manual or rule-based partition and junction identification methods with automated algorithms that process depth maps and point clouds. This substitution uses computational geometry and pattern recognition algorithms to automatically detect line segments, intersections, and building partitions, achieving high measurement precision without user interaction
2Productivity
If manual segmentation methods are used, then user interaction can correct errors, but the process is time-consuming and less accurate
Solution Approach 1:
The patent performs preliminary processing of building scenes by generating depth maps from point clouds and pre-processing the data into standardized formats before segmentation. This preliminary action prepares the data in advance with appropriate transformations and filtering, enabling faster subsequent processing while maintaining accuracy through pre-computed geometric relationships
Solution Approach 2:
The patent implements self-service automation where the system automatically identifies partitions and junctions without requiring user interaction. The algorithms autonomously process depth maps, detect line segments, identify intersections, and generate segmentation results independently, achieving both high productivity and maintained precision through iterative refinement and validation
3Ease of manufacture
If simple training datasets are used for neural network training, then training is faster and easier, but the network performance degrades in complex scenes with noise and distractors
Solution Approach 1:
The patent performs preliminary data augmentation and preprocessing to create robust training datasets. Synthetic depth maps are generated with controlled additions of noise, distractors, and varying scene complexities before training. This preliminary preparation ensures the neural network encounters diverse conditions during training, improving reliability without requiring manual collection of complex real-world datasets
Solution Approach 2:
The patent systematically varies parameters in training datasets including noise levels, distractor densities, scene configurations, and depth map resolutions. By changing these parameters across multiple training iterations, the neural network learns to handle complex scenes robustly. The systematic parameter variation provides comprehensive coverage of possible scene variations while maintaining automated dataset generation
Data Source
AI summary
A computer-implemented method for segmenting a building scene including obtaining a training dataset of top-down depth maps. Each depth map includes labeled line segments and junctions between line segments. The method further includes learning, based on the training dataset, a neural network. The neural network is configured to take as input a top-down depth map of a building scene comprising building partitions and to output a scene wireframe including the partitions and junctions between the partitions. This constitutes an improved solution for scene segmentation.