Towards Collision Avoidance for UAVs to Guide the Visually Impaired
Abstract
In this poster, we explore a mechanism to detect obstacles within a distance ’d’ ahead of a visually impaired person (VIP) and offer them a path with a minimum width ‘w’ to navigate between the obstacles.
System Overview
We present an assistive navigation system using Unmanned Aerial Vehicles (UAVs) to support visually impaired persons (VIPs) in obstacle-rich environments. The UAV flies behind the VIP, passively tracking their movement while scanning the path ahead for obstacles within a predefined distance d.
Using onboard sensors, the UAV detects objects and identifies open pathways with a minimum required width w. This enables it to guide the VIP with timely feedback, helping them avoid collisions and stay within safe navigable zones.
Methodology
The onboard edge hardware, carried by the VIP, handles all processing to maintain low latency. The processing stack includes lightweight Convolutional Neural Networks (CNNs) to identify and localize objects in the field of view, along with depth perception to calculate the distance of these objects from the VIP.
Using this data, a path planning module computes feasible routes by analyzing gaps between obstacles and determining navigable regions.
Guidance cues are delivered to the VIP through audio or haptic feedback systems, offering intuitive, non-visual direction changes during navigation.
Outcomes
Initial experiments in controlled settings show that the system successfully maintains a safe buffer around obstacles and consistently identifies viable paths. It adapts effectively to cluttered environments while remaining compact and responsive.
This approach demonstrates the viability of UAV-based assistive mobility systems that are both autonomous and user-centric, offering a promising solution for real-world guidance of visually impaired individuals. Future work includes outdoor testing, dynamic obstacle handling, and longitudinal usability studies with real users.
