Andi Surya, Ida Wahyuni

Traffic Density Detection on Public CCTV of Malang City Government Using YOLOv8

Introduction

Traffic density detection on public cctv of malang city government using yolov8. Detect traffic density in Malang City using YOLOv8 on public CCTV data. This automated system achieves 92.8% accuracy, aiding congestion management on major roads.

Abstract

The increasing number of motor vehicles in Malang City has led to a rise in traffic volume, resulting in a higher risk of congestion, especially on major roads such as Jalan Ahmad Yani. To address this issue, an automated system capable of detecting and monitoring traffic density is needed. This study aims to develop a traffic density detection system using the YOLOv8 object detection model with video data from public CCTV operated by the Government of Malang City. The methodology includes video data collection, data preprocessing, model training using YOLOv8, model testing, and traffic density calculation based on the number of detected vehicles. The model was evaluated during training using precision and recall metrics, resulting in a precision of 89.4% and a recall of 85.2%. Model testing was conducted on test videos by calculating accuracy based on a comparison between the number of vehicles detected by the system and the manual vehicle count, resulting in an average accuracy of 92.8%. These results indicate that the model is capable of accurately detecting vehicles in real-world conditions. This system can serve as a foundation for developing automated traffic monitoring systems that utilize visual data from CCTV.

Review

This paper addresses a highly relevant and practical problem: urban traffic congestion, specifically in Malang City. The authors propose an automated system leveraging the state-of-the-art YOLOv8 object detection model to detect and monitor traffic density using readily available public CCTV footage. This approach offers a cost-effective and potentially scalable solution for traffic management, moving beyond manual observation. The reported performance metrics, including a precision of 89.4%, recall of 85.2%, and an average detection accuracy of 92.8% against manual counts, are encouraging and suggest the model's robust capability in identifying vehicles under real-world conditions. The use of actual public CCTV data is a significant strength, demonstrating the system's potential for immediate applicability. The methodology outlines a clear process from video data collection and preprocessing to model training and evaluation, utilizing standard object detection metrics. While the reported accuracy is commendable, the abstract could benefit from further elaboration on certain aspects. For instance, more details on the characteristics of the training and test video datasets, such as their duration, varying lighting conditions, and typical traffic volumes, would provide a richer context for the model's performance. Furthermore, the definition of "traffic density calculation" could be more explicit; is it purely a vehicle count, or does it incorporate spatial considerations or flow rates? Understanding the model's robustness regarding challenging scenarios like heavy occlusion, diverse vehicle types, or adverse weather conditions would also strengthen the analysis. Overall, this study presents a promising foundational step towards developing intelligent traffic monitoring systems that can significantly aid urban planning and congestion management. The integration of YOLOv8 with public CCTV data offers a practical and scalable solution for leveraging existing infrastructure. Future work could focus on extending this system to distinguish between different vehicle types (e.g., cars, motorcycles, trucks) for more granular analysis, incorporating temporal dynamics for traffic flow prediction, and exploring real-time deployment strategies. Further validation across a wider range of CCTV locations and environmental conditions, alongside an analysis of computational efficiency for large-scale deployment, would also enhance the paper's contribution.

Full Text

Comments

You need to be logged in to post a comment.