Work

Adaptive visual perception and navigation framework for safe indoor mobility of visually impaired users

Year: 2026
Type: article
Abstract: Indoor mobility for visually impaired people is an essential problem due to dynamic obstacles and inadequate real-time detection systems. Existing solutions have been found lacking in terms of robustness and adaptability. The current research develops a robust adaptive visual perception and navigation system based on obstacle detection and indoor navigation in real time. The system consists of You Only Look Once version 11 (YOLOv11) algorithm for obstacle detection, Multi-Scale Feature Enhancement (MSFE) for better representation of smaller and distant obstacles, a pixel-wise replacement approach through Kernel Density Estimation (Ps-KDE) for pre-processing with illumination variation, deep reinforcement learning (DRL) for adaptive decisions in navigation, and Adaptive Coyote-Badger Optimization (ACBO) for learning. The developed indoor high-resolution dataset consists of 23 obstacle classes and was used to train and evaluate the proposed algorithm; the NYU Depth v2 and SUN RGB-D datasets were also utilized for validation. This model has been able to generate 97.52% Mean Average Precision (mAP) and 96.67% Mean Average Recall (mAR) in the customized dataset with an average frame per second of 55.6 FPS, which makes it suitable for mobile applications. For the scale evaluation, distance estimation, camera position, ablation, and real-life tests are carried out to further evaluate each component as well as the entire framework. A mobile application for providing navigation guidance through text to speech has been implemented. This framework shows its effectiveness under different indoor scenarios without depth sensing.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7227710137
FWCI:
Cites: 2
Cited by:
Related to: 10