A Comprehensive and Systematic Review of Deep Learning Based Object Recognition Techniques for the Visually Impaired

Abstract

Object detection is one of computer vision's most essential and challenging visual recognition tasks. The primary focus is to detect objects belonging to different classes and assign each one of them a predefined class label. The dearth of support at diagnosis centers, restricted access to actions and data, stigma created by society, and lack of employment frequently lead visually challenged individuals to seclusion. Existing technologies include A Path Force Feedback Belt (PF belt), Eye Substitution, and Radio Frequency Identification Walking Stick (RFIWS). They detect and localize objects to offer the visually impaired an understanding of the environment around them utilizing the operations of sensors. These systems do not provide real-time responses and are suitable only for outdoor coverage. Due to ultrasonic or radio frequency limitations, the detection range of these systems is between 1m to 5m. The primary intent of this work is to deliver a complete and extensive analysis of the research done in the field of object recognition for VI-challenged people.