Floor Space Optimisation and Recommendation System in 2D Space

Abstract

In this paper, we are proposing an open-source solution for empty space calculation in images by segmenting out the objects using computer-vision technologies. We have pre-existing solutions for 3D floor mapping but there is no end-to-end approach that can be implemented in a 2D space. Our solution takes an input image of a scene like an empty balcony or a room and provides us a 2D floor mapping of the space. Using transfer learning, a custom trained semantic segmentation model is used to identify the objects in images. The outcome of semantic segmentation model is used by our custom algorithm to determine the empty floor space in an image and segment the area into restricted and non-restricted regions. The restricted segments, like the pathway in-front of the doors, furniture etc., are the ones that should be kept empty for people to move around easily. The non-restricted floor area can be utilised for a wide number of use-cases like design recommendations, autonomous robot navigation, empty parking space identification etc. This paper mainly deals with the balcony garden design recommendation, however as mentioned the same approach can be extended to other spaces.