学进去

While there are already apps that guide blind users to a bus stop’s approximate location, those people may unknowingly end up standing too far away from the actual stop. A new app addresses that shortcoming by letting the smartphone’s camera in on the act.

Known as All_Aboard, the AI-based app was developed by a team of scientists at the Harvard-affiliated Massachusetts Eye and Ear treatment and research center. It’s made to be used along with a third-party GPS-based online navigational (导航的) map.

Users start by using other apps to get to a bus stop’s location. They then open All _Aboard and proceed to hold their smartphone up so that its back camera can “see” the surrounding street. Drawing upon a deep learning neural network which was trained on around 10,000 images of bus stops within that city, the app is reportedly able to visually identify the target stop’s sign if it’s within a range of up to 15 meters away. Once the sign has been spotted, the app guides the user via sonar-like beeps (哔哔声) which change pitch and speed as the person gets closer to the bus stop.

In field tests, 24 legally blind volunteers used both online maps and All _Aboard to locate 20 bus stops—10 in an urban setting and 10 in a suburban environment.

When it came to getting close enough to those stops, online maps alone had a success rate of just 52%, whereas All _Aboard boosted the number to 93%. Additionally, while the average distance between the map’s endpoint and the actual bus stop was 6.62 meters with online maps, it was just 1.54 meters with All _Aboard.

“Our findings suggest that the All _Aboard app could help travelers with visual impairments (视力受损) 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 Massachusetts Eye and Ear’s Assoc. Prof. Gang Luo. “This study indicates that computer vision-based object recognition capabilities can be used in a complementary (补充的) way and provide added benefit to purely mapping-based, macro-navigation services in real-world settings.”