A hybrid object detection approach for visually impaired persons using pigeon-inspired optimization and deep learning...

Area of Science:

• Computer Science
• Artificial Intelligence
• Assistive Technology

Background:

• Visually impaired individuals face challenges navigating their environment due to limited visual perception.
• Existing technological solutions for the visually impaired often raise privacy concerns regarding location sharing.
• Deep learning (DL) offers potential for advanced object detection to aid daily living and safety.

Purpose of the Study:

• To develop a reliable and precise object detection system for visually impaired persons.
• To enhance the safety and effectiveness of navigation for visually impaired individuals through advanced technology.
• To propose a novel deep learning approach that overcomes limitations of existing assistive technologies.

Main Methods:

• A Hybrid Approach to Object Detection for Visually Impaired Persons Using Attention-Driven Deep Learning (HAODVIP-ADL) was developed.
• Image pre-processing utilized bilateral filtering (BF) for noise reduction and edge preservation.
• Object detection employed the YOLOv10 framework, with backbone fusion of CapsNet and InceptionV3 for feature extraction. Classification used a multi-head attention and bi-directional long short-term memory (MHA-BiLSTM) approach, optimized with pigeon-inspired optimization (PIO).

Main Results:

• The HAODVIP-ADL method demonstrated superior performance in object detection tasks.
• Experimental validation on the Indoor Objects Detection dataset yielded an accuracy of 99.74%.
• The approach effectively integrates diverse spatial and contextual information for accurate object recognition.

Conclusions:

• The proposed HAODVIP-ADL technique significantly improves object detection accuracy for visually impaired users.
• This advanced deep learning model offers a promising solution for enhancing the independence and safety of visually impaired individuals.
• The method provides a reliable and precise system for real-time environmental perception and navigation assistance.