Development and Evaluation of Collision Avoidance User Interface for Assistive Vision Impaired Navigation

In this paper, an initial novel user interface prototype is developed and evaluated to assist vision impaired people to avoid collision with moving people while navigating in indoor environments. The user interface performs both pose classification and distance estimation based on RGB images and sends the results to the decision tree classifier model which classifies whether it is safe for vision impaired person to navigate, or should the vision impaired person be cautious or stop to avoid collision with the moving person. Experimentation showed that the user interface with the combined performance of pose classification, distance estimation and decision tree model showed an accuracy of 93.55% on a testing video dataset.