Point clouds to direct indoor pedestrian pathfinding

Abstract

Increase in building complexity can cause difficulties orienting people, especially people with reduced mobility. This work presents a methodology to enable the direct use of indoor point clouds as navigable models for pathfinding. Input point cloud is classified in horizontal and vertical elements according to inclination of each point respect to n neighbour points. Points belonging to the main floor are detected by histogram application. Other floors at different heights and stairs are detected by analysing the proximity to the detected main floor. Then, point cloud regions classified as floor are rasterized to delimit navigable surface and occlusions are corrected by applying morphological operations assuming planarity and taking into account the existence of obstacles. Finally, point cloud of navigable floor is downsampled and structured in a grid. Remaining points are nodes to create navigable indoor graph. The methodology has been tested in two real case studies provided by the ISPRS benchmark on indoor modelling. A pathfinding algorithm is applied to generate routes and to verify the usability of generated graphs. Generated models and routes are coherent with selected motor skills because routes avoid obstacles and can cross areas of non-acquired data. The proposed methodology allows to use point clouds directly as navigation graphs, without an intermediate phase of generating parametric model of surfaces.