Revolutionizing Signage Analysis: Leveraging YOLOv7 Object Detection for Comprehensive Classification and Assessment of Diverse Signage Types
John Paul Q. Tomas, Adam Lee B. Buenaventura, Jester D. Cruzate, Ghasutt Joshua R. Patarata, Izak Kyle E. Villanueva · 2024
Signage plays an essential role in the industry, where it displays its form in various kinds of media advertisements, directional road signs, as well in the realms of digital spaces. This study aims to use YOLOv7 for object detection to classify different signages and validate the model compared to the ground truth through calculated evaluation. The researchers used a video walkthrough gathered from an online media platform that was then processed through an annotating website to categorize Normal, Light Emitting Diode, and Traffic signages. The dataset will then be preprocessed with traditional learning and deep learning models. Google Colaboratory will be used to program the model in question and the development and testing in classifying the various images. Data augmentation is one of the key recommendations that is inherited in the study proper as it is able to solve the imbalance that can be recognized in the data proper. In the results of the study, it was adhered that the means of detection for the classes normal and LED signages proved to have scores of higher than 60 percent in the metric formulas that were utilized, while for the classification of traffic signages proved to be minimal in classification scoring lower than 6 percent. Reasons for such can derive from minimal data significance of the particular class, resulting in the means of lesser identification and incoherence with the model proper in identification.