Akademik Veri Yönetim Sistemi
9th INTERNATIONAL CONFERENCE ON CARTOGRAPHY & GIS, Burgas, Bulgaristan, 16 - 21 Haziran 2024, ss.1-2
Indoor navigation
poses challenges due to obstacles like solid surfaces blocking GNSS signals and
the absence of indoor maps, particularly in large buildings like hospitals,
airports, and shopping malls. Additionally, the complexity of indoor spaces,
with varying architectures and multiple floors, complicates navigation. Users
need to rely on their mental maps during navigation, but the irregular layout
of indoor environments increases cognitive load and may lead to navigation
errors. In this study, a Natural Language Processing (NLP) pipeline extracts
qualitative spatial concepts from verbal route descriptions and decomposes them
into landmarks and actions. The experimental data consists of crowdsourced
route descriptions obtained from a campus building. The pipeline takes human
route descriptions for indoor routes as input. It segments sentences, tokenizes
words, performs part-of-speech tagging, and dependency parsing to label spatial
objects, explicit relationships, and actions. The phrases are sorted into
sequences integrating landmarks and actions into the route solution of the navigation
network. Generated metric-based instructions are compared to each human route
instruction and matching human route instruction is added to the route
guidance. Besides, the advantages and disadvantages of human route descriptions
in reducing cognitive load for indoor wayfinding compared to the metric-based
route descriptions are discussed.