Real-time Stairs Detection and Distance Estimation System Based on Android for Visually Impaired

Abstract

Visually impaired individuals face significant risks when navigating stairs due to limited or no vision. This research proposes a real-time Android-based system to detect stairs and estimate their distance to assist visually impaired users. The system integrates Single Shot Multibox Detector (SSD) with MobileNet for object detection and applies Pinhole camera model for distance estimation. Data is collected using a chest-mounted smartphone camera at a distance of 2 to 5 meters. The results showed that the system achieved high detection accuracy for “Stairs Up” with an average accuracy of $98.91 \%$ at $2 \mathrm{~m}, 99.24 \%$ at $3 \mathrm{~m}, 99.42 \%$ at 4 m, and $96.14 \%$ at 5 m. In contrast, the “Stairs Down” detection showed high detection accuracy for “Stairs Up” with an average accuracy of $98.91 \%$ at 2 m. In contrast, the “Stairs Down” detection showed an average accuracy of $84.63 \%$ at 2 m. The distance estimation error ranged from 0.08 to 0.24 meters. The system has the advantages of inexpensive and real-time staircase detection and auditory alert capabilities using only a smartphone, making it highly accessible. These results show that the proposed system can significantly improve mobility, awareness, and safety for visually impaired individuals.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint