UAVDT Dataset

Overview

The UAV Detection and Tracking dataset (UAVDT) is designed for object detection and tracking from drone perspectives in urban environments. It focuses primarily on vehicle detection and tracking.

Key Statistics

• 100 video sequences (~80,000 frames)
• Over 1 million annotated bounding boxes
• 3 object categories (car, truck, bus)
• Multiple weather conditions and camera movements
• Various altitudes (15-70 meters)

Dataset Attributes

• Weather Conditions: Daylight, night, fog
• Flying Altitudes: Low altitude (15-30m), medium altitude (30-50m), high altitude (50-70m)
• Camera Views: Front view, side view, bird view
• Vehicle Density: Sparse, moderate, dense
• Vehicle Scales: Small, medium, large

Strengths

• Rich attribute annotations (weather, altitude, etc.)
• Diverse camera movements
• Well-structured for benchmarking
• Realistic urban scenarios

Limitations

• Limited to vehicle classes
• Focused on urban environments
• Less diverse than VisDrone
• No non-vehicle annotations

Applications

• Urban traffic monitoring
• Vehicle counting and classification
• Traffic flow analysis
• Parking lot monitoring

Source

Dataset Page: https://paperswithcode.com/dataset/uavdt