Object Detection Algorithms for Autonomous Navigation Wheelchairs in Hospital Environment

Abstract

Under the overwhelming situation medical facilities struggle to provide sufficient personnel to assist the continuous traffic. Therefore, considering alternative solutions to help nurses transport patients is relevant. An autonomous navigation wheelchair driven by object detection can become a substitute for transporting patients between facilities. The objective of this research is to study the potential of object detection algorithms in facilitating the autonomous navigation of patients within hospitals. For an object detection model to be a standalone technology for driving the autonomous wheelchair it has to satisfy the standards of the domain and display adequate readiness for deployment. Precisely, conduct viability experimentation based on Efficiency, Safety, and Reliability on a You Only Learn One Representation (YoloR) model trained for the research purpose. The paper finds YoloR as an appropriate model with potential for deployment. Both in terms of Reliability and Safety, the algorithm fits the designated criteria. However, the Autonomous Wheelchair system is a real-time system with strict execution time requirements that YoloR on its own does not reach. The latter is further confirmed by the proposed testing method.