This study introduces three key datasets for drone detection, tracking, and 3D trajectory analysis, developed to overcome the limitations of existing public object detection datasets (e.g., ImageNet, MSCOCO), which lack sufficient drone imagery. These datasets include both 2D and 3D data to support comprehensive drone trajectory analysis in various scenarios.

• A 2D drone image dataset consisting of real and synthetic drone images

• Contains a total of 12,541 images with ground-truth bounding boxes

• Includes four drone models: Air2S, Mavic3, Mini3Pro, Tello

2DRONE (ON): 1,107 real images collected online

2DRONE (AUG): 11,434 synthetic images
- Air2S: 2,085 images
- Mavic3: 2,668 images
- Mini3Pro: 3,245 images
- Tello: 3,436 images

• A synthetic 3D drone trajectory dataset created using Blender

• Reflects the actual size of four drone models

• Resolution: 640×480, frame rate: 30fps

• Includes nine different scenarios for diverse 3D trajectory patterns:

- Seq. #01-03: A single drone moving linearly along a single axis
- Seq. #04-05: A scenario with two drones crossing paths
- Seq. #06-07: A single drone with nonlinear movement
- Seq. #08-09: Complex scenarios combining rotation and nonlinear movement

• A real-world drone dataset for validation in real environments, supporting 3D trajectory analysis

• Uses the Tello drone model

• Employs a stereo camera setup for 3D position calculation of the drone via binocular geometry

• Consists of two sequences:
  - Seq. #10-11