Chief Supervisor

Prof. Allen CHEONG

Project Title

Predicting navigation patterns of visually impaired patients based on machine learning and the applications in computer-aided navigation

Synopsis

Good orientation, navigation and wayfinding skills are important for independent travel. Unfortunately, these skills deteriorate as we age or become disabled, reducing older people’s mobility and community engagement. Because of reduced visual input due to vision loss, low vision patients may have difficulties to acquire accurate navigation and positioning efficiently, reducing their ability to travel independently. Hence, they have to rely on their caregivers or companions to travel, reducing their independency and physical activities. 

In recent decades, with significant increases in computing power and rapid advances in artificial intelligence technology, key technologies such as route navigation and obstacle avoidance have been available to assist people’s navigation.

>Current navigation aids have limitations for providing reliable and efficient recognition and detection, which cannot fully meet the travel requirements of the visually impaired in terms of experience, practicality, interactivity and safety. Hence, patients’ independent travel remains very difficult, which seriously affecting their quality of life.

Prior to designing a more user-friendly and reliable navigation aid, it is important to understand the navigation behavior adopted by visually impaired patients. For example, how low vision patients acquire their navigation positioning and reliable information at different scenarios during navigation. 

Due to the wide range of vision loss (from very mild to very severe low vision), patients may establish different navigation behavior and motor skills, which might be very different from their sighted peers. Their navigation strategies might be varied dependings on the environments. To analyze such large amount of behavior data, we shall use machine learning by simulating and building datasets for different scenarios (indoor, outdoor, daytime, nighttime), different people (congenital/ acquired, children/ adolescents/ middle-aged/old) and walking styles (speed, step length, up/down, arc). This database will provide more concrete information about common navigation modes adopted by patients. Adding personalized navigation mode applications, a more customized computer-aided algorithm will be developed which can predict walking patterns and modes for the visually impaired. This can truly help the visually impaired patients solve their access problems and minimize the machinization of previous assistive devices.