Based on the linked source; automatically prepared and checked against the original report.
Purpose and scope
A research record describes XMDF-WDDAM, a proposed system intended to identify wheelchair users and support accessibility monitoring in smart urban environments. Its purpose is to help assistive systems, rehabilitation professionals, facility managers, and infrastructure efforts assess mobility support and access. The source is an abstract summary, not a full-paper review.
Detection and feature fusion
The workflow begins with image-quality preprocessing, followed by a YOLOv9-based detector designed for real-time environments. ResNet 18, EfficientNetV2-S, and a Vision Transformer extract features, which are fused into a unified representation. An attention-based fully connected neural network learns patterns from that representation, while AdamW is used to improve training efficiency and convergence.
Reported evaluation
Grad-CAM++ visualises important regions in input images. The evaluation used the benchmark Wheel Chair Dataset and compared the method with state-of-the-art models. The record reports 99.43% precision in those simulations. Authors conclude it offers an accurate, explainable approach for automated detection and accessibility-aware monitoring, but the supplied material gives no broader deployment evidence.
