Towards modular intelligent design method of subway station spatial with PointNet+

Area of Science:

• Computer Vision
• Artificial Intelligence
• Architectural Design

Background:

• Subway station layouts exhibit similarity and replicability, making them suitable for intelligent modular design.
• Traditional design methods may not fully leverage the potential of 3D spatial data for optimizing functional layouts.
• PointNet++ is a deep learning architecture effective for object recognition and semantic segmentation in 3D point clouds.

Purpose of the Study:

• To develop and evaluate a deep learning-based modular design method for subway station functional spaces using 3D spatial data.
• To assess the efficacy of the PointNet++ network in recognizing and classifying architectural elements within subway station environments.
• To demonstrate the capability of the proposed method for efficient and accurate identification and classification of architectural spaces.

Main Methods:

• Collected 3D plane data from subway stations in multiple cities.
• Constructed 3D models and exported point cloud data (X, Y,Z, rgbC), followed by data enhancement.
• Trained the PointNet++ network on an 8:1:1 split training, verification, and test dataset.

Main Results:

• PointNet++ achieved stable training with a mean loss of 0.42 and training accuracy of 0.76.
• The model demonstrated strong performance on the test set with an average class accuracy of 0.80+ and an overall accuracy of 0.75+, and a mean IoU of approximately 60%.
• High correlation between predicted and ground truth results indicated the model's self-learning ability and capacity for scheme space optimization.

Conclusions:

• The proposed modular design method, leveraging deep learning on 3D spatial data, efficiently and accurately identifies and classifies architectural spaces.
• The PointNet++ network is well-suited for processing 3D subway station data, facilitating effective information transmission for design.
• This approach offers a significant advancement in the intelligent, modular design of functional spaces within complex architectural environments like subway stations.