Real-time Machine Vision System for the Visually Impaired

This work aims to provide a robust assistive system that uses the interface of a conversational chatbot to guide people who are blind in outdoor navigation. The proposed system will also be aware of situations where people close to the user are not wearing masks. We propose and analyze the effectiveness of a sophisticated system consisting of a mask detector, common objects detector and an obstacle avoidance system. The proposed system can be entirely hosted on a secure remote server and used on any smartphone with stable internet connectivity, with all the processing handled by the server. The responses each sub-system generates are transmitted back to the client, where the user listens to the response. The proposed system outperforms to detect masks with an accuracy of 99.51% and object detection with an 80.36% success rate, respectively.