aldo pamungkas, shevti arbekti, elliya sestri

Penerapan YOLO dan OpenCV dalam Klasifikasi Kendaraan pada Lalu Lintas Kota Depok

Introduction

Penerapan yolo dan opencv dalam klasifikasi kendaraan pada lalu lintas kota depok. Deteksi & klasifikasi kendaraan lalu lintas Depok menggunakan YOLOv8 dan OpenCV. Sistem real-time akurasi tinggi (mAP@0.5 91%) mendukung transportasi cerdas.

Abstract

The growth in the number of vehicles in Depok City has driven the need for an accurate and efficient traffic monitoring system. This study implements the You Only Look Once (YOLO) version 8 algorithm to automatically detect and classify vehicles based on Python and OpenCV. The focus of the study is on four types of vehicles, namely motorcycles, private cars, buses, and trucks. The dataset was obtained from CCTV recordings and field documentation, then annotated using LabelImg and processed into YOLO format. The training process was carried out using the pretrained YOLOv8 model, while the system testing was conducted on videos of Depok City roads. Model performance was evaluated using the metrics of mAP@0.5 and mAP@0.5:mAP95, precision, recall, and F1 score. The evaluation results show that the model achieved a mAP@0.5 of 91% and a mAP@0.5:mAP95 of 75.1%, precision of 88.5%, recall of 85.2%, and an F1-score of 86.8%. With these results, the model is capable of detecting and classifying vehicles in real time with high accuracy under various lighting conditions and camera angles. Additionally, this system is integrated with a web interface using Flask for direct visualization of detection results. This research contributes to supporting smart transportation systems in urban environments and provides a potential solution for data-based traffic management.

Review

This study presents a timely and relevant implementation of the YOLOv8 algorithm for vehicle classification in the urban traffic environment of Depok City. Addressing the growing need for efficient traffic monitoring, the authors leverage Python and OpenCV to develop a system capable of detecting and classifying motorcycles, private cars, buses, and trucks. The abstract highlights promising performance metrics, including a high mAP@0.5 and mAP@0.5:mAP95, indicating a robust and accurate model. The integration with a Flask-based web interface for direct visualization is a commendable practical feature, positioning this work as a valuable contribution to smart transportation systems and data-driven traffic management. The technical approach utilizing YOLOv8 is well-justified given its reputation for real-time object detection capabilities, and the use of a pretrained model for training is a standard and effective practice. The methodology for data acquisition, involving both CCTV recordings and field documentation, followed by annotation with LabelImg, suggests a diligent effort in preparing a relevant and representative dataset for the specific locale. The reported evaluation metrics—mAP@0.5 of 91%, mAP@0.5:mAP95 of 75.1%, precision of 88.5%, recall of 85.2%, and an F1-score of 86.8%—demonstrate a high level of accuracy and reliability. Crucially, the abstract claims the model's effectiveness under "various lighting conditions and camera angles," which is a critical aspect for real-world deployment of such a system. While the abstract paints a positive picture, a full paper would benefit from greater detail in several areas. For instance, information regarding the size and diversity of the dataset, including the number of images/videos and instances per class, would provide a clearer understanding of the model's training foundation. Discussion of specific challenges encountered during deployment in Depok City, such as occlusion, varying vehicle sizes, or unique local traffic patterns, would also enhance the study's depth. Future work could explore expanding the classification to include more granular vehicle types, integrating vehicle counting and speed estimation functionalities, or evaluating the system's performance under extreme weather conditions. Further elaboration on the computational resources required for "real-time" operation and a comparison with existing traffic monitoring solutions specific to Depok City or Indonesia would further solidify the practical contribution of this promising research.

Full Text

Comments

You need to be logged in to post a comment.