Evaluating the Performance of a Visual Support System for Driving Assistance using a Deep Learning Algorithm

Beg, Mohammad Sojon; Ismail, Muhammad Yusri; Miah, Md Saef Ullah · 2023 · OpenAlex-citations

DOI: 10.37934/araset.34.1.3850

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical issue of road safety in Malaysia, where high traffic volumes and diverse driving conditions contribute to significant accident rates. The authors aim to evaluate the effectiveness of a visual support system for driving assistance, specifically focusing on object detection in local road environments. The research targets the detection of three key classes: vehicles, motorcycles, and traffic lamps, using the YOLO V8 deep learning algorithm. This approach seeks to improve intelligent transportation systems by leveraging computer vision technologies tailored to the unique characteristics of Malaysian roads, including varied infrastructure and climate conditions. The methodology involved creating a custom dataset from video recordings captured along the route from Hospital Pekan to the University of Malaysia Pahang campus. Approximately 1,000 images were extracted, annotated, and managed using the Roboflow dataset manager. The dataset contained roughly 10,000 vehicle instances, 900 motorcycles, and 3,700 traffic lamps. The data was partitioned into 70% for training, 20% for testing, and 10% for validation. The YOLO V8 model was trained in a Google Colab environment, and its performance was assessed using precision, recall, F1 score, and mean Average Precision (mAP). The results indicate that the model achieved an overall F1 score of 71%, precision of 88.2%, and recall of 84% on the training set. Performance varied by class, with traffic lamps achieving the highest mAP of 0.844, followed by vehicles at 0.776, and motorcycles at 0.523. Validation results were comparable to training metrics, with an average F1 score of 0.72 and precision of 1.00 at an 88.3% confidence threshold. The model demonstrated strong generalization capabilities, though motorcycle detection remained less accurate than other classes. Visual analysis confirmed the system’s ability to localize objects in both straight road and junction scenarios. The study concludes that YOLO V8 is effective for detecting vehicles and traffic lamps in Malaysian road conditions, offering a viable solution for enhancing driver assistance systems. However, the lower performance in motorcycle detection highlights a need for further optimization. The findings suggest that locally relevant datasets are crucial for developing robust object detection models in developing countries. Future work should focus on improving motorcycle detection accuracy and expanding the scope to include additional object classes and road scenarios to create a more comprehensive traffic monitoring system.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• dms validation