A Real-Time Assistive Vision System using YOLOv5

Assistive technologies have become increasingly prominent over the past few years because of their extensive usage. There have been many studies conducted over the last two decades on technological implementation into medicine and the development of assistive technologies geared toward those individuals with disabilities. The implemented technologies may assist individuals with visual impairments with navigation and object detection. The proposed system's software incorporates real-time object recognition, optical character recognition, and speech synthesis capabilities utilizing a single-stage, modified deep learning object detection model designed specifically for real-time object recognition and accurate object identification without latency or privacy concerns. Image input is taken from a live camera feed, the input image is processed, and the object recognition results are provided based on performance data with a low latency rating and privacy level; therefore, the final product is subject to the determined embedded constraints from the tested architecture. The test environment was both controlled (data set) and uncontrolled (i.e., living environment), and data collected during testing was used to evaluate the product's effectiveness and determine the suitability of using low-cost technology to create a vision-based assistive system to field-test products.