Faster R-CNN with Equilibrium Optimizer for Real-Time Object Detection and Classification Framework to Aid Visually Impaired People

Abstract

In general, blind or Visually Impaired People (VIP) survive their lives in hardship, with the main challenge that they are incapable of living entirely independently. Object detection is a major contribution of computer vision and Machine Learning (ML) in the detection of objects in videos or images. Numerous research works have been implemented in the field of real-time object detection utilizing ML and Deep Learning (DL). This study focused on a precise and real-world object detection method for visually challenged persons, developing a novel Equilibrium Optimizer-based Real-Time Object Detection and Classification Framework to support VIP (EORTODCF-VIP). The image preprocessing stage applies a Mean Filter (MF) to remove noise. The Faster-RCNN model is employed for object detection to efficiently generate specific region proposals and identify objects within images. Then, the InceptionResNetV2 framework is implemented for the backbone of the feature extractor method, and a Wavelet Neural Network (WNN) is applied for object detection and classification. To improve the classification performance of the WNN classifier, hyperparameter tuning is performed through the Equilibrium Optimizer (EO) method. The effectiveness of the EORTODCF-VIP technique was evaluated on a benchmark image dataset, demonstrating greater performance compared to existing methods.