Inaudible echo-based indoor position recognition

• Main Author: Wong, Alexander Chong Xan
• Other Authors: Tan Rui
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/148126
• Institution: Nanyang Technological University

There has been a growing need for accurate location positioning using mobile devices. The most widely used system for localisation today is the Global Positioning System (GPS). However, GPS lacks the accuracy required for precise indoor location triangulation. An example where accurate indoor localisation would be useful is contact tracing to limit the spread of the SARS-CoV-2 virus amid the Covid-19 pandemic. Prior Research has been conducted in this field, the most noteworthy is RoomRecognise. With an accuracy if up to 99.7%, RoomRecognise has pushed the frontier of accurate indoor localisation. However, RoomRecognise is not perfect, the paper has established that there are decreases in the accuracy of their approach when the orientation of the mobile device is changed. This paper presents two approaches with the goal of accurate indoor localisation at all orientations. The approaches build upon the foundation set by the RoomRecognise system in pursuit of this goal. The approach with the best results is the Multi-Channel approach contained within this paper. Unlike the model created by RoomRecognise, this approach combines the RoomRecognise model with another Multi-Layer Perceptron into a single unified large neural network in order to process both audio data and orientation data simultaneously. Tests conducted have shown that the Multi-Channel approach has achieved better performance than the RoomRecognise Model with an accuracy of 92% vs 79%, thus achieving the goal of designing an accurate indoor localisation approach for all orientations.