AI smartphone app helps visually impaired commuters find bus stops

A team of researchers from Mass Eye and Ear have developed a micro-navigation smartphone app to provide assistance to those who are blind or visually impaired (BVI) in finding their bus stops.

Current GPS systems have sufficient macro-navigation for planning routes using public transportation. However, micro-navigation, such as finding the exact locations of bus stops and destinations, remains an issue for people who are BVI, as GPS-based localisation for this is less accurate. 

To combat this problem, the researchers developed a mobile app called All_Aboard, which is meant to be used in conjunction with mainstream GPS systems and focuses on improving micro-navigation. 

When a GPS indicates that a BVI user is nearing their destination, that is when All_Aboard should be opened. The app uses the phone’s camera to detect street signs from 30-50ft away. It then uses auditory cues to direct the user towards their destination, with the frequency of the sounds changing as they approach the end point.

The app is powered by artificial intelligence, using a deep learning neural network trained on about 10,000 images of bus stops collected in a given city or region. The app is currently capable of recognising bus stops in 10 major cities/regions around the world (this video demonstrates how the app works).

In the study, 24 BVI individuals used All_Aboard along with Google Maps to navigate a set route with 10 bus stops at an urban (Boston) and suburban site (Newton, Massachusetts). The results of the study were measured in terms of localisation error and rate of successful localisation. 

Localisation error, or gap distance, is defined as the distance between the desired destination and the map's marked end point. The rate of successful localisation is the probability of getting close enough to the bus stops. 

The researchers found that in both urban and suburban locations, All_Aboard had a success rate of 93 percent, whereas Google Maps had a 52 percent success rate. Additionally, the average gap distance with Google Maps was 6.62m and 1.54m with All_Aboard.

GPS accuracy is supposed to be acceptable in suburban areas, according to Luo, who added it was not initially expected that the performance with Google Maps in Newton, Mass. would be so low. Previous research from Luo and his team found this problem may be due to widespread errors in bus stop location mapping data in Google Maps.

“Our findings suggest that the All_Aboard app could help travellers with visual impairments in navigation by accurately detecting the bus stop, and therefore greatly reducing their chance of missing buses due to standing too far from the bus stops,” said Gang Luo, PhD, of the Schepens Eye Research Institute of Mass Eye and Ear. 

“This study indicates that computer vision-based object recognition capabilities can be used in a complementary way and provide added benefit to purely mapping-based, macro-navigation services in real-world settings.