Intelligent Assistant System Based on Single Camera and Deep Neural Networks for Aiding Visual Impaired Individuals

Identifier:  TFM:946

Handle: https://hdl.handle.net/20.500.11797/TFM946

Authors:  Farouk Marei, David George

Abstract:
Object Detection and Depth Estimation methods can help visual impaired individuals to understand the scene in front of them. there are multiple applications that provide help to those individuals by connecting them (by a video call) to others who can help in describing the scene to them. However, We believe that I can give an alternative to those applications using light and fast machine learning models and avoid the interaction of the human support. The project is divided into Two parts. The First part is Object Detection in a scene. We used YOLOV5S model that was trained on the COCO date-set with 80 different objects. The second part in the depth estimation model. We used Midas from pytorch after trying multiple depth models. the depth estimation model will help me to estimate the distance of each object extracted from the depth model. The final goal is to take an image or a video from the user and extract the objects with the distance of each of them and send this data to the user in a sound note format.