AI-Assisted 3D Building Models from Raster Plans with Semantic Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The loss of structured geometric and semantic information during the conversion of 2D construction plans from vector to raster formats hinders the ability to generate accurate 3D models, as essential details like room information, height, and hidden structural elements are difficult to recover.
Innovation Solution
Utilizing artificial intelligence, specifically an artificial neural network (ANN) trained on building standards and practices, to predict and extrapolate missing semantic information from rasterized 2D plans, converting them back to vector format with integrated semantic data to construct a 3D model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If 2D construction plans are converted from vector to raster format for storage and sharing, then ease of operation and portability are improved, but structured geometric and semantic information is lost
Solution Approach 1:
The system performs preliminary extraction of structured geometric and semantic information from vector plans before rasterization occurs. This captured data is stored separately and can be recovered later, preventing information loss that would otherwise occur during the vector-to-raster conversion process.
Solution Approach 2:
The patent introduces an intermediary data structure that bridges between raster images and 3D models. This intermediary captures semantic information from the original vector plans and preserves it in a format that can be used for 3D reconstruction, acting as a mediator that prevents information loss during format conversion.
2Ease of operation
If rasterized 2D plans are used for digital storage, then ease of operation is improved, but manufacturing precision of 3D model generation deteriorates
Solution Approach 1:
The system extracts and stores structured geometric data and semantic information from vector plans before rasterization. This preliminary capture of precise measurement data ensures that when 3D models are generated later, the original precision is recovered even though the working format is raster-based.
Solution Approach 2:
The patent replaces traditional mechanical measurement methods (manually measuring from raster images) with an automated AI-based system that retrieves precise geometric data from the stored structured information, substituting manual processes with intelligent automated reconstruction.
3Device complexity
If simple raster format is used for plan storage, then device complexity is reduced, but ability to detect and measure hidden structural elements deteriorates
Solution Approach 1:
The patent introduces an intermediary semantic data layer that captures information about hidden structural elements (electrical, plumbing, HVAC) during the vector-to-raster conversion process. This intermediary data structure allows the simple raster format to carry additional semantic information that enables detection and measurement of hidden elements without increasing the complexity of the storage system.
Data Source
AI summary
Systems and methods allow prediction of semantic information and generation of a 3D model from one or more raster images of a structure. In embodiments, the raster images may be converted to vector images, and received semantic information may be used for generation of predicted semantic information. The combination of provided and predicted semantic information, along with the vector images, can facilitate generation of a 3D model that is scale-accurate to the structure illustrated by the raster images. Other embodiments may be described and/or claimed.