Room Classification in Floor Plan Recognition

Abstract

Home improvement companies usually build 3D models based on 2D floor plans, and then the computer will automatically simulate the decoration of different rooms, so clarifying the exact type of room becomes a necessary task for the automatic decoration. Although it is possible to recognize text in floor plans, and the technology for text recognition is well developed, the challenge is to get the type of room when the floor plan lacks a textual description of the room. When text is missing from the floor plan, it means that you need to guess the type of room. To solve this problem, this paper attempts to apply the classification algorithm to solve the problem of rooms lacking textual information and to complete the classification of the rooms using the improved C4.5 algorithm. To enable the classification task to be performed smoothly, this paper also proposes a door and window detection method based on the YOLOv3 model. This method can accurately identify doors and windows, so the generated vector graphics can become more accurate, which is beneficial to the classification task. The experiments show that the YOLOv3 model-based door and window detection method achieves good results on the recognition task, and the mAP (mean Average Precision) of this model is 94.013% on the dataset we build. On the other hand, the improved C4.5 algorithm also enables efficient room classification with an accuracy of 78.71% and it is 6.22% more accurate than the traditional C4.5 algorithm.