Enhanced Real-Time Highway Object Detection for Construction Zone Safety Using YOLOv8s-MTAM

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Object detection is critical for autonomous driving safety, especially in dynamic highway construction zones.
• Existing systems struggle with high-speed, occluded, or blurred objects common in these environments.

Purpose of the Study:

• To develop an enhanced YOLOv8s object detection system with improved robustness for autonomous driving in construction zones.
• To integrate a motion-temporal attention module (MTAM) for better detection of dynamic and occluded objects.

Main Methods:

• An enhanced YOLOv8s architecture incorporating a CSP backbone, FPN-PAN feature fusion, and advanced loss functions.
• Integration of a novel motion-temporal attention module (MTAM) utilizing temporal convolutions and attention mechanisms.
• Training on a custom dataset of 34,240 images with extensive data augmentation and 9-Mosaic transformations.

Main Results:

• Achieved high performance metrics: mAP(IoU[0.5]) of 90.77 ± 0.68% and mAP(IoU[0.5:0.95]) of 70.20 ± 0.33%.
• Demonstrated strong real-world recognition rates: 96% for construction vehicles, 92% for warning signs, and 84% for flag bearers.
• The MTAM effectively improved detection of blurred and partially occluded objects.

Conclusions:

• The enhanced YOLOv8s system with MTAM significantly improves object detection reliability in challenging highway construction zones.
• The framework shows strong potential for real-time deployment in intelligent transportation systems for enhanced safety.