An intelligent optimized object detection system for disabled people using advanced deep learning models with...

Area of Science:

• Computer Vision and Artificial Intelligence
• Assistive Technologies for Disabilities

Background:

• Visually impaired individuals face daily challenges that technology can address.
• Object detection (OD) using computer vision (CV) and machine learning (ML) offers potential for enhanced navigation and problem recognition for the visually impaired.
• Deep learning (DL) models trained on diverse datasets show promise for developing specialized assistive technologies.

Purpose of the Study:

• To propose a novel Object Detection System for Disabled People Using Advanced Deep Learning Models and Sparrow Search Optimization (ODSDP-ADLMSSO).
• To improve the object detection capabilities for visually impaired persons (VIPs).
• To enhance secure and informative navigation for VIPs through advanced AI.

Main Methods:

• Image preprocessing using a Gaussian filter (GF) to reduce noise and improve clarity.
• Object detection (OD) performed using the YOLOv7 model for identification, localization, and classification.
• Feature extraction via the MobileNetV3 model and classification using a temporal convolutional network (TCN).
• Hyperparameter optimization for the TCN model implemented with the sparrow search optimization algorithm (SSOA).

Main Results:

• The ODSDP-ADLMSSO method was evaluated on the Indoor OD dataset.
• The proposed system achieved a superior accuracy of 99.57%.
• Demonstrated significant performance improvement over existing object detection techniques.

Conclusions:

• The ODSDP-ADLMSSO approach effectively enhances object detection for visually impaired individuals.
• The integration of advanced DL models and optimization algorithms leads to highly accurate assistive technology.
• This system holds potential for improving the safety and independence of visually impaired users through better navigation.