Vehicle Counting Based on Vehicle Detection and Tracking from Aerial Videos.

Area of Science:

• Computer Vision
• Traffic Engineering
• Robotics

Background:

• Unmanned aerial vehicles (UAVs) offer flexible deployment and a wide perspective for traffic monitoring.
• Traditional sensors have limitations in coverage and deployment flexibility compared to UAVs.
• Accurate vehicle counting from aerial imagery is crucial for effective traffic management.

Purpose of the Study:

• To propose a novel framework for accurate vehicle counting from UAV-captured aerial videos.
• To address challenges posed by static and moving backgrounds in aerial traffic scenes.
• To enhance the robustness of vehicle detection and tracking under varying conditions.

Main Methods:

• A moving-object detector framework handling both static and moving backgrounds.
• Pixel-level foreground detection with continuous background model updates for static backgrounds.
• Image registration for camera motion estimation and vehicle detection in a reference system for moving backgrounds.
• Online-learning tracker to manage changes in vehicle scale and shape.
• Multi-object management module utilizing multi-threading for efficient analysis.

Main Results:

• The proposed framework demonstrated high accuracy in vehicle counting from aerial videos.
• Achieved over 90% accuracy for fixed-background videos.
• Achieved over 85% accuracy for moving-background videos.
• The method effectively handles variations in vehicle scale and shape.

Conclusions:

• The developed framework provides an effective solution for vehicle counting using UAVs.
• The method's ability to handle diverse background conditions enhances its applicability in real-world traffic monitoring.
• This research contributes to advancing intelligent transportation systems through aerial video analysis.