Extracting Key Traffic Parameters from UAV Video with On-Board Vehicle Data Validation

Area of Science:

• Transportation Engineering
• Computer Vision
• Remote Sensing

Background:

• Unmanned Aerial Vehicle (UAV) video offers advantages in flexibility and data richness for traffic monitoring.
• Existing methods require systematic approaches for accurate vehicle detection and parameter extraction.

Purpose of the Study:

• To develop and validate a systematic approach for vehicle detection and tracking using UAV video.
• To extract key traffic parameters with high accuracy.

Main Methods:

• Utilized the YOLO v3 model for object detection and the DeepSORT algorithm for vehicle tracking.
• Conducted field experiments with UAVs at varying heights (150-500m) and speeds (40-90 km/h).
• Collected 5400 frames and 1192 speed points using GNSS-RTK and onboard OBD for validation.

Main Results:

• Achieved 90.88% accuracy in object detection and 98.9% precision in vehicle tracking.
• Extracted vehicle speeds with an absolute error within ±3 km/h and a relative error within 2%.
• Demonstrated an overall accuracy of 98% for extracted traffic parameters.

Conclusions:

• The proposed UAV-based approach is robust and reliable for traffic monitoring.
• The system effectively detects, tracks vehicles, and extracts critical traffic data.
• This method enhances the capabilities of intelligent transportation systems.