Indoor Signage and Arrow-Icon-Text Detection and Localization with YOLOv5

Approximately 2.2 billion people worldwide live with vision impairment, posing considerable challenges to their daily navigation. While tools like walking sticks help them avoid obstacles, they fall short in offering precise directions and navigation, particularly in indoor environments. Improving assistive solutions can enable visually impaired individuals to move independently and lessen their dependency on others. This paper involves the use of YOLOv5 for indoor Signage and Arrow-Icon-Text (AIT) detection and localization. Two custom models are generated, one for signage and the other for AIT detection and localization. These models use a custom dataset with 750 real-world images comprising 1300 signages, 1214 arrows, 1372 icons, and 2999 texts. Experimental testing shows promising results with a mean Average Precision (mAP50) of 90.7% for Signage Detection and 97.6% for AIT Detection.