Semantic Segmentation and Topological Mapping of Floor Plans

Abstract

When visually impaired people walk in an unknown indoor environment, it is crucial to build a topological semantic map from the captured floor plan for navigation purposes. This paper proposes a topological mapping method from the floor plan model based on deep learning semantic segmentation. The topological semantic map can be used for assistive blind navigation purposes in unknown indoor environments. A deep learning network is developed for semantic segmentation, and disturbances such as image rotation, color transformation and Gaussian noises are taken into consideration in the training to enhance the robustness. With the semantic segmentation result as input, a topological semantic mapping algorithm is then proposed based on the graph theory. Experiments are presented to demonstrate the effectiveness of the proposed mapping method.