TrackAid-DT: vision for visually impaired athletes
An egocentric vision dataset and lane detection benchmarks to support independent running for visually impaired athletes.
Visually impaired runners usually depend on a sighted guide to stay in their lane. TrackAid-DT is a step towards assistive systems that can do this automatically: it provides an egocentric (first-person) vision dataset recorded on athletics tracks, together with benchmarks for lane detection models.
Contribution
- A new egocentric dataset that captures the track from the runner’s point of view, in daylight, at night, in low light and during sprints.
- Benchmarks of lane detection methods on this data, establishing a baseline for assistive guidance systems.
- A deployment study on edge devices and a wearable haptic glove that turns lane deviations into vibration cues.
From data collection to guidance: segmentation models, edge deployment and haptic feedback. Graphical abstract from Narang et al. (2026), Computer Vision and Image Understanding, CC BY 4.0.
The wearable setup: a chest-mounted camera and a glove with vibration motors and a BLE microcontroller. Figure from Narang et al. (2026), Computer Vision and Image Understanding, CC BY 4.0.
The work is published in Computer Vision and Image Understanding (Narang et al., 2026). I was the first author.
References
2026
- CVIU
TrackAid-DT: An egocentric vision dataset and benchmarks for lane detection in visually impaired athlete guidanceComputer Vision and Image Understanding, 2026
