Obayya Marwa, Al-Wesabi Fahd N, Alshammeri Menwa, Iskandar Huda G
Department of Biomedical Engineering, College of Engineering, Princess Nourah bint Abdulrahman University, P.O. Box 84428, Riyadh, 11671, Saudi Arabia.
Department of Computer Science, Applied College at Mahayil, King Khalid University, Mahayil Aseer, Saudi Arabia.
Sci Rep. 2025 May 13;15(1):16514. doi: 10.1038/s41598-025-00608-z.
Visually impaired persons face several problems in their day-to-day lives, and technological intermediaries might help them encounter their challenges. Among other beneficial technologies, object detection (OD) is a computer technology related to image processing and computer vision (CV), which identifies and describes objects like vehicles, animals, and persons from digital videos and images. Visually impaired persons (VIPs) can utilize the OD approach for detecting problems and recognizing services to offer secure and informative navigation. Recently, machine learning (ML) and deep learning (DL) have been trained with numerous images of objects, which are highly related to people with disabilities. In this article, a novel Object Detection System for Disabled People Using Advanced Deep Learning Models and Sparrow Search Optimization (ODSDP-ADLMSSO) approach is proposed. The main aim of the ODSDP-ADLMSSO model is to enhance the OD method for visually challenged people. At first, the Gaussian filter (GF) is employed in the image pre-processing stage to remove noise and make the image input data more transparent. In addition, the YOLOv7 method is used for the process of OD to identify, locate, and classify objects within an image. Furthermore, the MobileNetV3 model is utilized for the feature extraction process. The temporal convolutional network (TCN) model is implemented for classification. Finally, the hyperparameter selection of the TCN model is implemented by the sparrow search optimization algorithm (SSOA) model. The efficiency of the ODSDP-ADLMSSO method is examined under the Indoor OD dataset. The comparison study of the ODSDP-ADLMSSO method demonstrated a superior accuracy value of 99.57% over existing techniques.
视障人士在日常生活中面临诸多问题，而技术媒介或许能帮助他们应对这些挑战。在其他有益技术中，目标检测（OD）是一种与图像处理和计算机视觉（CV）相关的计算机技术，它能从数字视频和图像中识别并描述诸如车辆、动物和人物等物体。视障人士（VIP）可利用目标检测方法来发现问题并识别服务，以提供安全且信息丰富的导航。近来，机器学习（ML）和深度学习（DL）已通过大量与残疾人密切相关的物体图像进行了训练。在本文中，提出了一种新颖的使用先进深度学习模型和麻雀搜索优化的残疾人目标检测系统（ODSDP - ADLMSSO）方法。ODSDP - ADLMSSO模型的主要目标是增强针对视障人士的目标检测方法。首先，在图像预处理阶段采用高斯滤波器（GF）去除噪声，使图像输入数据更清晰。此外，使用YOLOv7方法进行目标检测过程，以识别、定位和分类图像中的物体。再者，利用MobileNetV3模型进行特征提取过程。实施时间卷积网络（TCN）模型进行分类。最后，通过麻雀搜索优化算法（SSOA）模型实现TCN模型的超参数选择。在室内目标检测数据集下检验了ODSDP - ADLMSSO方法的效率。ODSDP - ADLMSSO方法的比较研究表明，其准确率高达99.57%，优于现有技术。
