Real-Time Path Hole Detection for Motorcycle Driver Safety Assistance Using Faster Object More Object (FOMO) Algorithm
John Louise R. Ronda, Ian Edward DG. Pangan, Cristine Jeremie A. Nicolas, Desiree Jhen G. Sales, Paul Ryan A. Santiago, Reagan L. Galvez, Dennis R. dela Cruz, Oliver R. Mariano · 2024
This paper presents a real-time pothole detection system using the Faster Objects, More Objects (FOMO) object detection algorithm deployed on an embedded vision camera leveraging TinyML technology. The system uses the TensorFlow Lite Deep Learning framework to train the model on the color depth of RGB and grayscale images and evaluate its hyperparameter configurations to determine the best-performing pothole detection model. A thorough model analysis and evaluation achieved an F1 score of 81% in both the training and testing phases, indicating balanced model prediction performance. Despite current limitations in hardware processing specifications, this paper emphasizes the potential of TinyML technology in leveraging vision-based road safety features for various applications.