VisioGuide: Modified YoloV7 with ResNext for Object Detection to Aid Visually Impaired Individuals

People with visual impairment face multiple unique challenges in navigating and interacting with their surroundings. As the demand for revolutionary assistive technology grows to improve the movability of visually impaired people, Deep learning has emerged as a key solution. The need of revolutionary assistive technology made to improve the mobility and independence of visually disabled people is increasing by each passing day. Using the YOLOv7 object detection algorithm along with a few modifications in its backbone, Visioguide deploys deep learning algorithms to accurately detect and identify real-time objects, landmarks, and points of interest in real-time, thereby enabling seamless navigation and heightened spatial awareness for its users. The enhancements in YoloV7 backbone using weight initialization and spatial attention have increased the efficiency of object detection for visually impaired individuals. This robust technology serves as an indispensable aid, thereby making the visually impaired feel a part of the society and helping them to perform their daily chores. The paper proposes an efficient approach to object detection using modified YoloV7 along with voice feedback to aid visually impaired individual. The model is effective and convenient to detect general objects with mAP value of 59.2%. The paper proposes a study of YoloV7 object detection using modified ResNext Backbone Architecture and voice feedback of detected class.