Yus - A Deep Learning Algorithm for Collision Avoidance through Object and Vehicle Detection

Mohammad Sojon Beg; Muhammad Yusri Ismail; M. Saef Ullah Miah; Mohamad Heerwan Peeie · 2023 · Crossref

DOI: 10.37934/araset.31.1.226236

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 need for enhanced driving assistance systems to mitigate road accidents in Malaysia, a region with high rates of traffic fatalities caused by reckless driving and infrastructure challenges. The research specifically investigates the application of deep learning for real-time object detection, aiming to identify vehicles, motorcycles, and traffic signals in local road conditions. The motivation stems from the limitations of traditional computer vision algorithms, which struggle with occlusion, weather variations, and high-speed tracking, as well as the potential performance gaps of existing models when deployed in environments with unique local signage and road infrastructure. The methodology employed the YOLO-V8 deep learning algorithm for object detection. Data was collected using a dash-mounted camera capturing video footage of traffic on roads near Universiti Malaysia Pahang, including junctions and straight lanes, with the vehicle traveling at speeds between 60 and 80 km/h. The resulting dataset comprised 1,068 annotated images labeled via the Roboflow platform. This dataset was partitioned into training (70%), testing (20%), and validation (10%) subsets. The model was trained and evaluated in a Google Colab environment, utilizing metrics such as mean average precision (mAP), recall, precision, and confusion matrices to assess performance. The system architecture involved image preprocessing, feature extraction via convolutional neural networks, and bounding box prediction using anchor boxes. The results demonstrated that the YOLO-V8 model effectively detected the target objects with high accuracy. On the training dataset, the model achieved a mean average precision (mAP) of 88.2% and a recall of 71%. Validation results showed an mAP of 88.3%, with specific metrics of 70.1% for mAP50 and 30.5% for mAP50-95. Confusion matrix analysis indicated strong classification performance, particularly for motorcycles (0.50 actual prediction rate, 0.05 false prediction rate) and vehicles (0.75 actual prediction rate, 0.06 false prediction rate). Traffic lamps achieved a high actual prediction rate of 0.85, though with a higher false prediction rate of 0.17. The system exhibited low inference times, processing images with 4.3ms inference latency, indicating suitability for real-time applications. The significance of this work lies in its validation of YOLO-V8 as a robust tool for intelligent transportation systems in Malaysian contexts. By successfully detecting critical road objects under local conditions, the study supports the development of collision avoidance systems that can alert drivers to potential hazards. The findings suggest that deep learning models can overcome traditional detection limitations, offering a pathway to improve road safety and reduce accident rates. The research highlights the importance of fine-tuning algorithms with local data to address specific environmental challenges, contributing to the broader field of autonomous driving and traffic management.

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.

• distraction detection algorithms