Automatic Traffic Sign, Animal Detection, and Recognition Using YOLOv8n to Avoid Human-Animal Road Conflicts
Abstract
This study presents a deep learning-based solution to mitigate human-animal conflicts on the road, utilizing YOLOv8n for Automatic Traffic Sign, Animal Detection, and Recognition (ATSADR). In addition, this study introduces a custom dataset that includes Indian traffic signs and animal images (i.e., monkey and deer) collected from the Indian Institute of Technology Madras (IITM) campus. Furthermore, the collected dataset is used to train the YOLOv8n model (with and without pre-trained weights). A detailed comparative analysis of the performance metrics such as: recall, mean Average Precision (mAP), accuracy, precision, F1-Score, and computational efficiency has been performed to select the most suitable model for the specified task. Real-time testing has been conducted to demonstrate the practical performance of the chosen model. The findings signify the efficacy of the trained YOLOv8n model in real-time scenarios, showing promising solutions for addressing human-animal roads conflicts. The trained model achieved a mAP of 73.7% and an average accuracy of 99.3%. The present work holds great potential to help prevent road accidents involving animals (deer and monkeys), and if trained with different animal datasets, the YOLOv8n model will offer a proactive solution to mitigate human-animal conflicts, ultimately enhancing the overall safety of both drivers and wildlife.