Indoor positioning systems (IPS) are commonly used in robotics and essential for many autonomous operations as GPS can be unreliable indoors and is, therefore, infeasible for such applications. In this project, we lay the groundwork for a pedestrian positioning system, which can map indoor spaces and track pedestrian migrations inside buildings using mobile phones. The IPS revolves around using known methods in robotics, such as simultaneous localization and mapping(SLAM), to see if they can be adapted to work using only the sensors available on mobile phones. Research into phone sensors and mainly the accelerometer will be done to determine if the sensor values returned by a modern mobile phone are up to the task of mapping pedestrian movements accurately. This will be done by plotting the accelerometer’s movements to see how accurate they are and creating a machine-learning algorithm to classify movement patterns such as walking and running. The results from testing the sensors did not match what would be expected, likely due to the algorithms used by the mobile phone. Using the raw sensor data and creating an algorithm might give different results. Classifying movements worked with around 92-95% accuracy, but due to COVID-19, gathering data became very hard, so there might be over-fitting in the results. This research mainly shows what does not work when creating an indoor positioning system using mobile phones rather than what does. This research shows that using a mobile phone positioning system to create an IPS has limits. This report eliminates many options and does not find the ultimate solution. But it is hopefully a step towards a pedestrian SLAM. We will contrast this problem with its counterpart in robotics, discuss the added complexities, and propose possible solutions to mitigate them.
People
Baldur Þór Haraldsson
Egill Torfason