Treffer: Intelligent feature fusion with dynamic graph convolutional recurrent network for robust object detection to assist individuals with disabilities in a smart Iot edge-cloud environment.

Smart Internet of Things (IoT)-edge-cloud computing defines intelligent systems where IoT devices create data at the network's edge, which is then further processed and analyzed in local edge devices before transmission to the cloud for deeper insights and storage. Visual impairment, like blindness, has a deep effect on a person's psychological and cognitive functions. So, the use of assistive models can help mitigate the adverse effects and improve the quality of life for individuals who are blind. Much current research mainly concentrates on mobility, navigation, and object detection (OD) in smart devices and advanced technologies for visually challenged people. OD is a vital feature of computer vision that includes categorizing objects within an image, allowing applications like augmented reality, image retrieval, etc. Recently, deep learning (DL) models have emerged as an excellent technique for mining feature representation from data, primarily due to significant developments in OD. The DL model is well-trained with manifold images of objects that are highly applicable to visually impaired individuals. This paper presents an intelligent Feature Fusion with Dynamic Graph Convolutional Recurrent Network for Robust Object Detection (FFDGCRN-ROD) approach to assist individuals with disabilities. The paper aims to present an intelligent OD framework for individuals with disabilities utilizing a smart IoT edge cloud environment to enable monitoring and assistive decision-making. At first, the image pre-processing phase involves resizing, normalization, and image enhancement to eliminate the noise and enhance the image quality. For the OD process, the FFDGCRN-ROD approach employs the faster R-CNN to identify and locate specific targets within the images automatically. Furthermore, the fusion models, namely CapsNet, SqueezeNet, and Inceptionv3, are used for the feature extraction process. Finally, the FFDGCRN-ROD model implements the dynamic adaptive graph convolutional recurrent network (DA-GCRN) model to detect and classify objects for visually impaired people accurately. The experimental validation of the FFDGCRN-ROD methodology is performed under the Indoor OD dataset. The comparison analysis of the FFDGCRN-ROD methodology demonstrated a superior accuracy value of 99.65% over existing techniques.
(© 2025. The Author(s).)

Declarations. Competing interests: The authors declare no competing interests.