An Interactive Floor Plan Image Retrieval Framework Based on Structural Features

To build a house, buyers meet with an architect, who analyses the land’s built-up area and creates a development blueprint. Customers may have different perspectives on their residence. As a result, in order to save architects’ time and provide an appropriate floor plan to the customer, we presented an interactive framework that allows the client to change existing designs and search for similar ones in a repository. To do this, we created a vocabulary of annotations for inserting new items, deleting existing objects, shifting the position of objects, and moving the locations of objects. The proposed framework accepts a digital form of a printed floor plan image with or without hand-drawn annotations as a query, then performs suggested editing operations and converts the image to standard form by identifying annotations. The retrieval model then analyses the processed image and provides further recommendations similar to the query floor plan image. Our contributions include (1) an interactive platform for updating an existing floor plan based on predefined vocabulary of annotations, and (2) a unique structure-based features for retrieving similar floor plans with a \(10\%\) improvement in mAP score. The suggested framework’s performance was examined using publicly available datasets, and it obtained 0.85 mAP which edge over existing state-of-the-art approaches.