Related Experiment Video
Updated: Sep 3, 2026
Deep-Learning Based Multi-Joint Synchronous Tracking for Objective Quantification of Hindlimb Locomotor Kinematics in Rats
Published on: April 3, 2026
Explainable multi-backbone deep feature fusion based wheelchair detection with disability-aware accessibility
Mahmoud Ragab1, Ali Altalbe1, Alanoud Subahi2
1Information Technology Department, Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah, Saudi Arabia.
Background:
Reliable identification of wheelchair users in public environments is an important enabling technology for assistive systems that aids independent mobility, accessibility monitoring, and inclusive public infrastructure. Accurate automated recognition could assist rehabilitation professionals, facility managers, and smart accessibility systems in assessing mobility support and enhancing access for individuals with disabilities.
Objective:
This manuscript introduces an Explainable Multi-backbone Deep Feature Fusion based Wheelchair Detection and Disability-Aware Accessibility Monitoring (XMDF-WDDAM) approach. The major goal of the proposed model is to design an efficient wheelchair detection system for smart urban environments, classifying individuals into three categories: person, wheelchair, and not wheelchair.
Methods:
Towards this objective, the proposed model initially undergoes a preprocessing stage to enhance image quality. Subsequently, a YOLOv9-based object detection approach is employed to detect wheelchair users in real-time environments. For feature extraction, the proposed model employs ResNet 18, EfficientNetV2-S, and Vision Transformer, which capture diverse informative features from the input images. Then, the extracted features are fused to form a unified representation. Besides, an attention-based fully connected neural network is applied to learn important patterns from the fused representations. The AdamW optimiser is further applied to enhance training efficiency and accelerate convergence speed. Eventually, the Grad-CAM++ is leveraged for visualising the important regions in the input images.
Results:
The experimental result analysis of the XMDF-WDDAM methodology is performed utilising the benchmark Wheel Chair Dataset. The simulation results reported the superior outcomes of the XMDF-WDDAM technique with a precision of 99.43% over state-of-the-art models. Conclusion: The proposed XMDF-WDDAM framework provides an accurate and explainable solution for automated wheelchair detection and accessibility-aware monitoring in public environments.