Object Detection Using Artificial Intelligence in Autonomous Vehicles

Abstract

This research paper focuses on object detection, such as bicycles, motorcycles, persons, traffic lights, traffics signs, and vehicles within the framework of autonomous driving systems in the CARLA environment. Currently, object detection in autonomous driving primarily relies on actual autonomous vehicles, which face challenges such as high costs and real-time implementation difficulties. The open-source CARLA system enables precise and cost-effective experimentation. In this paper, the deep learning model YOLOv5 was used, yielding good results in both training and validation datasets. A total of 1560 different images were used in the training process, divided into 1120 images for training, 160 images for testing, and 320 images for validation. The training results showed a Precision (P) of 0.898, Recall (R) of 0.827, mAP@50 of 0.900, and mAP@50-95 of 0.583. In the validation results, the Precision (P) was 0.891, Recall (R) was 0.801, mAP@50 was 0.880, and mAP@50-95 was 0.542. These results indicate that the model is capable of accurately detecting and retrieving objects effectively.