Lightweight YOLOv5-Based Algorithm to Detect Room Nameplates for Autonomous Smart Wheelchair

About 10% individuals with visual impairment also use wheelchair, which makes them difficult to ambulate into a room that can only be distinguished by text independently. One thing that could be a solution is to implement a room nameplate recognition system on autonomous smart wheelchairs so that it can ambulate the user to the dedicated room autonomously. We proposed a lightweight YOLOv5-based method to detect room nameplates that is more suitable for embedded devices such as smart wheelchair. We improved YOLOv5 with ghost module and coordinate attention to reduce model complexity while still maintain the detection accuracy. Using room nameplate images data that has been collected by our self, the proposed method has 29 % less parameter with about the same accuracy as the original YOLOv5. With a low complexity of object detection model, our study may be utilized for room nameplate recognition system on smart wheelchair.