FVCap: An Approach to Understand Scanned Floor Plan Images Using Deep Learning and its Applications

Understanding the layouts and features of indoor scenes necessitates graphic recognition in scanned floor plan images. Scanning floor plans usually results in obvious and unanticipated interruptions. Accurately identifying symbols in such a scenario has long been a contentious issue in the document image analysis field. The task was performed with respectable efficiency using classic machine learning approaches such as hand-crafted features and deep neural networks. The strengths of the previously proposed Capsule network and VGG19 are combined in this research to create a hybrid network. The proposed network FVCap employs the VGG19 network for feature encoding and the Capsule network for classification and decoding from floor plan region images. The proposed approach can recognise and identify region elements in scanned floor plan images, as well as perform end-to-end room labeling with significantly more accuracy than other hybrid networks and conventional machine learning models.