While the current version of the LiDAR-based floor plan extraction system has successfully achieved its core objectives, the development process has highlighted several areas for enhancement. These opportunities—ranging from algorithmic improvements to expanded usability—will drive the system closer to professional-grade deployment.

Current Limitations

One limitation of the existing implementation is its reliance on rule-based segmentation. Although the RANSAC and DBSCAN algorithms have proven effective in detecting wall segments, their performance is sensitive to parameter tuning and the quality of the point cloud. In cluttered or irregular environments, these methods may underperform or fail to distinguish between structural and non-structural features.

Another constraint is the lack of real-time feedback during the scanning or processing stages. Users must complete an entire scan before assessing data quality or identifying occluded regions. This makes the workflow less adaptive and could result in repeated scanning.

Proposed Enhancements

1. Machine Learning Integration
   Integrating deep learning for wall detection could reduce reliance on hand-tuned heuristics. A CNN model trained on labelled indoor point cloud datasets could distinguish walls, furniture, and voids with higher confidence.

2. SLAM-Based Live Mapping
   Real-time SLAM (Simultaneous Localisation and Mapping) integration would provide feedback during data collection, helping users identify under-scanned areas immediately. This is especially useful in large or irregularly shaped buildings.

3. Expanded Format Support
   Future versions could support IFC (Industry Foundation Classes), SVG, and JSON-LD to enhance interoperability with BIM tools and web-based platforms.

4. Interactive Editing Tools
   Adding basic editing features to the GUI would allow users to manually adjust geometry, relabel segments, or delete outliers, offering greater control over final outputs.