BBeep: A Sonic Collision Avoidance System for Blind Travellers and Nearby Pedestrians
Seita Kayukawa, Keita Higuchi, João Guerreiro, Shigeo Morishima, Yoichi Sato, Kris Kitani, Chieko Asakawa · 2019 · Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (CHI '19) · doi:10.1145/3290605.3300282
Summary
This paper presents BBeep, an assistive suitcase-shaped system designed to help blind travellers navigate crowded public spaces such as international airports, train stations, and shopping malls. The authors argue that existing blind-navigation research has focused largely on static obstacles and has assumed that sighted pedestrians will clear the path for a blind user once they notice them. In practice, pedestrians are often distracted by smartphones, conversations, or facing away, so collisions remain common. Rather than notifying only the blind user about nearby people, BBeep emits audible beep alerts that are heard by both the user and surrounding pedestrians, prompting the crowd to clear the path in the manner of an airport cart or reversing truck. The system uses a ZED stereo camera mounted on the suitcase to capture RGB and depth data, detects pedestrians in real time using a YOLOv2 convolutional neural network, tracks them across frames, and extrapolates their likely trajectories. A warning sound is triggered only when a predicted trajectory is expected to cross an 'emergency line' in front of the user within a set time window, keeping emissions infrequent enough to remain socially acceptable. The authors first ran a corridor observation study of 399 pedestrian trajectories to determine which sound types (high, middle, low urgency) and emission timings (5 s vs 2.5 s ahead) most effectively deflected pedestrians, then designed a three-stage policy (low-urgency, intermediate-urgency, stop) and evaluated the complete system in a real-world airport user study.
Key findings
In the corridor study, the timing of sound emission had a statistically significant effect on pedestrian behaviour while the sound type did not: 5-second-ahead alerts produced greater minimal distances between pedestrians and the suitcase than 2.5-second alerts (p = 0.001), and any alert produced more clearance than silence (p = 0.03). Urgency level alone showed no significant effect, so the authors chose lower-urgency tones to limit annoyance. In the airport study, six blind participants navigated crowded gate areas using three interfaces: BBeep's external speaker, a bone-conducting headset delivering identical audio only to the user, and no sound. The speaker condition produced significantly fewer pedestrians with an imminent collision risk than the headset condition (mean 0.41 vs 2.00, p = 0.005) and a lower risk-continuity ratio (p = 0.009), indicating that broadcasting to pedestrians was more effective than notifying the blind user alone. Participants reported that sighted people noticed them and moved aside when the speaker was used, whereas the headset required them to verbally ask people to move. Participants judged the speaker appropriate for airports, train stations, and malls but inappropriate for quiet settings such as hospitals or libraries.
Relevance
BBeep reframes assistive navigation as a two-party problem — blind users and sighted pedestrians share responsibility for collision avoidance — and shows that externalising feedback to the crowd can outperform private notifications in dense environments. For accessibility practitioners, the study offers design principles applicable well beyond suitcases: that social acceptance of alerts depends heavily on emission frequency and context, that timing (advance warning) often matters more than the sound itself, and that assistive technology should acknowledge the behaviour of bystanders rather than idealising them. The work also demonstrates that real-time computer-vision pipelines are now deployable on mobility aids. Limitations include the small participant pool (n = 6), a focus on straight-line routes without wayfinding, and unexamined acceptability from the perspective of sighted pedestrians whose behaviour is being manipulated. Future work should integrate BBeep with turn-by-turn indoor navigation and explore quieter or more informative feedback modalities.
Tags: blind navigation · obstacle avoidance · collision avoidance · assistive technology · wearable technology · computer vision · pedestrian detection · sonification · auditory interface · orientation and mobility · indoor navigation · blindness and low vision