Lane-specific speed analysis in urban work zones with computer vision

Area of Science:

• Civil Engineering
• Computer Vision
• Traffic Safety

Background:

• Work zone speed is critical for road construction safety.
• Existing methods for speed measurement may be costly or impractical.
• Need for accurate, lane-specific speed data in work zones.

Purpose of the Study:

• To develop a computer vision (CV) technique for measuring lane-specific vehicle speeds in work zones.
• To analyze the influence of traffic control, lane position, and construction activity on work zone speeds.
• To provide data supporting road construction safety management.

Main Methods:

• Combined YOLOv5 object detection and Deep-SORT tracking algorithms to monitor vehicles.
• Collected 21 days of video data from traffic monitoring cameras.
• Developed a novel technique to infer construction activity presence based on object detection.
• Compared vehicle speeds with and without construction activity.

Main Results:

• The CV framework achieved speed measurement errors between 0-6.4 KPH.
• Traffic control with median barrels reduced average speeds by 15 KPH.
• Work zone adjacent lanes showed higher speed variations.
• Construction activity significantly impacted slow traffic speeds (p-value 0.01-0.03), but not fast traffic.

Conclusions:

• The CV technique offers a reliable and cost-effective solution for measuring work zone speeds.
• The derived data enhances detailed safety analysis in construction zones.
• The technique is also applicable for general traffic speed monitoring.