2022 - CONVR Conference: An Automatic Method of Information Extraction From 2D Floorplans
Date:
The 2D floorplan can provide lots of useful information such as geometric information related to the design of spaces and text information related to the annotations. Manually extracting information from 2D floorplans is normally time-consuming and error-prone. To address this issue, there have been studies using different kinds of machine learning-based recognition methods. However, these methods are normally applied on high-quality floorplans with little noise. The results are not good enough when it comes to floorplans with only hardcopy or image formats. Additionally, the training process requires a large quantity of data and a long training time. An automatic approach is proposed in this paper to help extract geometric parameters and text information from 2D floorplans. Firstly, the floorplan is denoised by eliminating the grid lines. Next, the geometric information is extracted based on the segmentation of spaces in the floorplan using morphology operations. The text information is recognized from the annotations in each space using the OCR algorithm. Finally, the geometric parameters of the segmented rooms are estimated with an average accuracy of 92.18%. The text information contained in each room can also be successfully extracted with an accuracy of 93.61%. The proposed method can be applied to automatically extract useful information from 2D floorplans with high accuracy. 
