Robust Path Back-Tracing Guidance System for Blind People
Indoor positioning and navigation is an active area of research due to the widespread use of devices equipped with micro-machined electromechanical systems (MEMS) sensors such as gyroscopes, accelerometers, and magnetometers. The MEMS inertial sensors and other popular technologies can provide information to determine the position and orientation of a person relative to a known initial location. A system that is able to produce accurate location data may find multiple applications in location-based services (LBS), safe wayfinding, and other related fields. For instance, a guidance system could be used to provide travel-related information to passengers as they use public transportation, or provide safe navigational directions to people who find themselves in unfamiliar environments. The research presented in this dissertation focuses on building robust systems that provide real-time and reliable travel-related information to help blind or visually impaired travelers reach a destination safely.
In this dissertation, I present two novel navigation systems and an openly accessible and annotated data set of inertial sensor time series data collected from blind people. The first system conveys travel-related information to blind passengers when using public transportation. The system makes use of pre-configured Wi-Fi access points placed in public transit vehicles and at bus stops to convey real-time, multi-modal travel-related information to any passenger, directly on his or her own smart device. The second more complex system helps blind people retrace the path taken inside a building and walk safely back to an initial location. This is ideal for situations in which a blind person is able to reach a certain location, for example with the assistance of a sighted guide, and needs to find his or her way back to the initial location. This robust path back-tracing guidance system is comprised of a turn detector based on a hidden Markov model (HMM) to robustly detect turns even in the presence of drift in the inertial measurements and noticeable body sway during gait, a step counter that uses filtered inertial sensor data to determine the number of steps walked along a path, and a path matching algorithm to track the user’s location. The step counter and turn detection models were trained on sensor data from WeAllWalk, an openly accessible and annotated data set of inertial sensor time series data created from blind walkers using a long cane or a guide dog. This system runs on a smart device and provides the guidance necessary to help a blind person retrace a path. A robust guidance system that supports safe blind wayfinding opens the doors to many blind travelers who would like to travel and move independently as they explore unfamiliar environments.