We provide a dataset for target detection and tracking in aerial imagery-M3OT, a multimodal vehicle detection and tracking dataset acquired by two Unmanned Aerial Vehicles（UAVs) in a high altitude region. The dataset consists of both RGB and infrared thermal (IR) modalities, with two drones` altitudes ranging from 100m to 120m. The dataset consists of 21,580 frames extracted from 8 hours of video, 10790 paired RGB- Infrared thermal (IR) images from two UAVs, and 220,000 bounding boxes across various environments, including suburban, urban, daytime, dusk, and night. This dataset can serve as a benchmark for object detection, multiple object tracking, and other computer vision tasks.

The approach begins with UAV-mounted multi-modal sensors, including RGB cameras, infrared sensors, and LiDAR, to ensure reliable data collection across diverse operational conditions. To train and evaluate the proposed model, the UAV Small Object Detection Dataset is utilized, which includes aerial imagery labeled across 10 categories of small and weak objects. This dataset comprises 717 training samples, 84 validation samples, and 43 test samples, enabling robust model development under real-world aerial scenarios.

Dataset Structure and Splits This particular version of the dataset is meticulously organized 80% of Training and 20% of Testing into standard splits to facilitate structured deep learning experimentation, ensuring unbiased model evaluation and comparability of results:

Training Set: Comprising 717 samples, this subset is dedicated to training deep learning models. It contains the largest portion of the data, allowing models to learn diverse patterns, features, and contextual information necessary for small object detection. Each image in the training set is accompanied by precise bounding box annotations for the target objects. Validation Set: Consisting of 84 samples, this set is used during the model development phase for hyperparameter tuning, model selection, and early stopping. It provides an unbiased estimate of the model's performance on unseen data during training, preventing overfitting to the training set. Test Set: With 43 samples, this independent set is reserved exclusively for the final evaluation of the trained models. Performance metrics derived from the test set are considered the most reliable indicator of a model's generalization capability to real-world, completely unseen data.

Multi Drone Detection

## Overview

Multi Drone Detection is a dataset for object detection tasks - it contains Drone annotations for 2,194 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

The dataset contains UAV footage of wild antelopes (blackbucks) in grassland habitats. It can be mainly used for two tasks: Multi-object tracking (MOT) and Re-Identification (Re-ID). We provide annotations for the position of animals in each frame, allowing us to offer very long videos (up to 3 min) completely annotated while maintaining the identity of each animal in the video. The Re-ID dataset offers two videos, that capture the movement of some animals simultaneously from two different UAVs. The Re-ID task is to find the same individual in two videos taken simultaneously from a slightly different perspective. The relevant paper will be published in the NeurIPS 2024 Dataset and Benchmarking Track. https://nips.cc/virtual/2024/poster/97563 Resolution: 5.4 K MOT: 12 videos ( MOT17 Format) Re-ID: 6 sets (each with a pair of drones) (Custom) Detection: 320 Images (COCO, YOLO)

• (1) Single Object Tracking (SOT)
  • (1-1) Train/Validation/Test
  • (1-2) Train/Validation
• (2) Multi-Object Tracking (MOT)
  • (2-1) Train/Validation
• (3) Re-Identification (ReID)
  • (3-1) Full Bounding Box Train/Validation
  • (3-2) 1/10 Bounding Box Train/Validation

All datasets are derived from the https://zenodo.org/records/15103888" target="_blank" rel="noopener">official release of the 4th Anti-UAV Challenge, featuring thermal infrared videos.

The VisDrone Dataset is a comprehensive benchmark developed by the AISKYEYE team at the Lab of Machine Learning and Data Mining, Tianjin University, China. Designed for various computer vision tasks associated with drone-based image and video analysis, the dataset serves as an essential resource for researchers and practitioners in the field.

Key Features

• Extensive Collection: The dataset comprises 288 video clips containing 261,908 frames and 10,209 static images, all captured using different drone-mounted cameras. This extensive collection showcases a wide range of environments, objects, and scenarios.
• Diverse Environments: VisDrone encompasses images and videos from 14 cities across China, covering both urban and rural settings. This diversity enhances the dataset's applicability to various real-world applications.
• Varied Object Categories: The dataset features a rich array of object categories, including pedestrians, vehicles, bicycles, and tricycles. This variety allows for robust training and evaluation of models across multiple object detection tasks.
• High-Quality Annotations: With over 2.6 million manually annotated bounding boxes, the VisDrone Dataset provides detailed ground truth data for object detection, tracking, and crowd counting tasks. Annotations also include attributes such as scene visibility, object class, and occlusion, enabling researchers to develop more effective models.

Dataset Structure

The VisDrone dataset is organized into five main subsets, each targeting a specific task:

• Task 1: Object Detection in Images
• Task 2: Object Detection in Videos
• Task 3: Single-Object Tracking
• Task 4: Multi-Object Tracking
• Task 5: Crowd Counting This structured approach facilitates focused training and evaluation for distinct computer vision challenges.

Applications

The VisDrone Dataset is widely used for training and evaluating deep learning models in various drone-based computer vision tasks, including:

• Object Detection: Identifying and localizing multiple object classes in images and videos.
• Object Tracking: Following individual objects across frames in video sequences, enabling applications in surveillance and traffic monitoring.
• Crowd Counting: Estimating the number of individuals in crowded scenes, which is valuable for urban planning and safety assessments.

Conclusion

The VisDrone Dataset stands out as a significant contribution to the field of drone-based computer vision. Its diverse sensor data, extensive annotations, and various task-focused subsets make it a valuable resource for advancing research and development in drone applications. Whether for academic research or practical implementations, the VisDrone Dataset is instrumental in fostering innovation in the rapidly evolving domain of drone technology.

The UAV low-altitude dataset for multi-target detection and tracking provides a specialized benchmark designed to support research in object detection and multi-object tracking under low-altitude unmanned aerial vehicle (UAV) surveillance scenarios. The dataset contains annotated images and video sequences of pedestrians, vehicles, and other common ground targets captured from low-flying UAV platforms, reflecting challenges such as small object size, frequent occlusions, scale variations, and complex backgrounds. All samples are labeled with bounding boxes and identity information to facilitate both detection and tracking tasks. This dataset aims to advance the development of lightweight and robust algorithms for real-time UAV-based monitoring applications, including public safety, traffic management, and intelligent surveillance.

This dataset contains UAV RGB videos (MP4) recorded with a Phantom4 RTK in a vineyard during the harvesting campaign of 2023. It also includes frames and annotations (PNG) to boost Object Detection and Tracking of grape bunches. There are two types of videos: (1) videos capturing the side of the canopy from a frontal point of view only, and (2) videos that collect the data from multiple perspectives to avoid leaf occlusion, common in commercial vineyards. All flights were executed 3 meters above ground level, with a clear sky and wind speed below 0.5 m/s.

Drone Video AI Object Tracking for Search Market Outlook

According to our latest research, the global Drone Video AI Object Tracking for Search market size reached USD 1.18 billion in 2024, and is projected to grow at a robust CAGR of 19.6% from 2025 to 2033. By the end of 2033, the market is forecasted to reach USD 5.50 billion. This impressive growth is primarily driven by the increasing adoption of advanced AI-driven drone technologies in critical search operations, enabling real-time object detection and tracking across diverse environments.

The primary growth factor fueling the Drone Video AI Object Tracking for Search market is the rising demand for rapid, accurate, and scalable search solutions in emergency and high-risk scenarios. As natural disasters, urban emergencies, and missing person incidents become more frequent, organizations are turning to drones equipped with AI object tracking to augment traditional search methods. These AI-powered drones can scan vast and challenging terrains, identifying objects of interest with remarkable speed and precision. The integration of deep learning algorithms and computer vision technologies has significantly improved the efficacy of search missions, reducing human error and response times. Furthermore, the ability of drones to operate in hazardous or inaccessible environments enhances safety for human personnel, making them indispensable tools for modern search and rescue operations.

Another key driver for market expansion is the increased investment by government and defense agencies worldwide. Public safety authorities are prioritizing the deployment of AI-enabled drone systems for surveillance, disaster response, and border security. The proliferation of high-resolution cameras, thermal imaging, and advanced sensors has made drones more versatile and effective in object tracking applications. Additionally, regulatory frameworks in several countries are evolving to support the safe integration of drones into national airspace, further accelerating market adoption. The synergy between public sector initiatives and private sector innovation is fostering a dynamic ecosystem where AI object tracking drones are rapidly becoming standard equipment for search and monitoring tasks.

Technological advancements in AI algorithms, edge computing, and cloud-based data processing are also transforming the landscape of drone video analytics. The continuous improvement of neural networks and real-time data transmission capabilities allows for more sophisticated object recognition and tracking, even in complex or cluttered environments. Cloud-based solutions enable collaborative search efforts, where data from multiple drones can be aggregated, analyzed, and visualized in real time. This technological leap is not only enhancing operational efficiency but also opening new avenues for commercial enterprises and environmental organizations to leverage drone video AI object tracking for diverse applications such as wildlife monitoring, infrastructure inspection, and environmental conservation.

Regionally, North America holds the largest share of the Drone Video AI Object Tracking for Search market, accounting for approximately 38% of the global market in 2024. This dominance is attributed to substantial investments in R&D, a strong presence of leading technology firms, and proactive government policies supporting drone integration. Europe follows closely, driven by stringent safety regulations and growing adoption in public safety and environmental monitoring. The Asia Pacific region is witnessing the fastest growth, with a projected CAGR of 22.1% during the forecast period, fueled by rapid urbanization, increasing disaster management needs, and supportive regulatory frameworks. Latin America and the Middle East & Africa are also emerging as promising markets, with growing interest in leveraging drone AI technologies for public safety and environmental applications.

Component Analysis

The Drone Video AI Object Tracking for Search market is segmented by

MMOT: The First Challenging Benchmark for Drone-based Multispectral Multi-Object Tracking

The MMOT dataset was presented in the paper MMOT: The First Challenging Benchmark for Drone-based Multispectral Multi-Object Tracking. The official code and further details can be found on the GitHub repository: https://github.com/Annzstbl/MMOT.

  Introduction

MMOT is the first large-scale benchmark for drone-based multispectral multi-object tracking (MOT). It integrates spectral… See the full description on the dataset page: https://huggingface.co/datasets/Annzstbl/MMOT.

Overview

The Songdo Vision dataset provides high-resolution (4K, 3840×2160 pixels) RGB images annotated with categorized axis-aligned bounding boxes (BBs) for vehicle detection from a high-altitude bird’s-eye view (BeV) perspective. Captured over Songdo International Business District, South Korea, this dataset consists of 5,419 annotated video frames, featuring approximately 300,000 vehicle instances categorized into four classes:

• Car (including vans and light-duty vehicles)
• Bus
• Truck
• Motorcycle

This dataset can serve as a benchmark for aerial vehicle detection, supporting research and real-world applications in intelligent transportation systems, traffic monitoring, and aerial vision-based mobility analytics. It was developed in the context of a multi-drone experiment aimed at enhancing geo-referenced vehicle trajectory extraction.

📌 Citation: If you use this dataset in your work, kindly acknowledge it by citing the following article:

Robert Fonod, Haechan Cho, Hwasoo Yeo, Nikolas Geroliminis (2025). Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery, Transportation Research Part C: Emerging Technologies, vol. 178, 105205. DOI: 10.1016/j.trc.2025.105205.

🔗 Related dataset: For precisely georeferenced vehicle trajectories extracted from the same large-scale multi-drone experiment, see Songdo Traffic: 10.5281/zenodo.13828384.

Motivation

Publicly available datasets for aerial vehicle detection often exhibit limitations such as:

• Non-BeV perspectives with varying angles and distortions
• Inconsistent annotation quality, with loose or missing bounding boxes
• Lower-resolution imagery, reducing detection accuracy, particularly for smaller vehicles
• Lack of annotation detail, especially for motorcycles in dense urban scenes with complex backgrounds

To address these challenges, Songdo Vision provides high-quality human-annotated bounding boxes, with machine learning assistance used to enhance efficiency and consistency. This ensures accurate and reliable ground truth for training and evaluating detection models.

Dataset Composition

The dataset is randomly split into training (80%) and test (20%) subsets:

A subset of 5,274 frames was randomly sampled from drone video sequences, while an additional 145 frames were carefully selected to represent challenging cases, such as motorcycles at pedestrian crossings, in bicycle lanes, near traffic light poles, and around other distinctive road markers where they may blend into the urban environment.

Data Collection

The dataset was collected as part of a collaborative multi-drone experiment conducted by KAIST and EPFL in Songdo, South Korea, from October 4–7, 2022.

• A fleet of 10 drones monitored 20 busy intersections, executing advanced flight plans to optimize coverage.
• 4K (3840×2160) RGB video footage was recorded at 29.97 FPS from altitudes of 140–150 meters.
• Each drone flew 10 sessions per day, covering peak morning and afternoon periods.
• The experiment resulted in 12TB of 4K raw video data.

More details on the experimental setup and data processing pipeline are available in [1].

Bounding Box Annotations & Formats

Annotations were generated using a semi-automated object detection annotation process in Azure ML Studio, leveraging machine learning-assisted bounding box detection with human verification to ensure precision.

Each annotated frame includes categorized, axis-aligned bounding boxes, stored in three widely-used formats:

1. COCO JSON format

• Single annotation file per dataset subset (i.e., one for training, one for testing).
• Contains metadata such as image dimensions, bounding box coordinates, and class labels.
• Example snippet:

{
 "images": [{"id": 1, "file_name": "0001.jpg", "width": 3840, "height": 2160}],
 "annotations": [{"id": 1, "image_id": 1, "category_id": 2, "bbox": [500, 600, 200, 50], "area": 10000, "iscrowd": 0}],
 "categories": [
  {"id": 1, "name": "car"}, {"id": 2, "name": "bus"},
  {"id": 3, "name": "truck"}, {"id": 4, "name": "motorcycle"}
 ]
}

2. YOLO TXT format

• One annotation file per image, following the format:

• Bounding box values are normalized to [0,1], with the origin at the top-left corner.
• Example snippet:

0 0.52 0.63 0.10 0.05 # Car bounding box
2 0.25 0.40 0.15 0.08 # Truck bounding box

3. Pascal VOC XML format

• One annotation file per image, structured in XML.
• Contains image properties and absolute pixel coordinates for each bounding box.
• Example snippet:

File Structure

The dataset is provided as two compressed archives:

1. Training Data (train.zip, 12.91 GB)

train/
│── coco_annotations.json # COCO format
│── images/
│  ├── 0001.jpg
│  ├── ...
│── labels/
│  ├── 0001.txt # YOLO format
│  ├── 0001.xml # Pascal VOC format
│  ├── ...

2. Testing Data (test.zip, 3.22 GB)

test/
│── coco_annotations.json
│── images/
│  ├── 00027.jpg
│  ├── ...
│── labels/
│  ├── 00027.txt
│  ├── 00027.xml
│  ├── ...

Additional Files

• README.md – Dataset documentation (this description)
• LICENSE.txt – Creative Commons Attribution 4.0 License
• names.txt – Class names (one per line)
• data.yaml – Example YOLO configuration file for training/testing

Acknowledgments

In addition to the funding sources listed in the metadata, the creators express their gratitude to Artem Vasilev for his dedicated efforts in data annotation. We also thank the research teams of Prof. Simon Oh (Korea University) and Prof. Minju Park (Hannam University) for their assistance during the data collection campaign, including the provision of drone equipment and student support.

Citation & Attribution

Preferred Citation: If you use Songdo Vision for any purpose, whether academic research, commercial applications, open-source projects, or benchmarking efforts, please cite our accompanying article [1]:

Robert Fonod, Haechan Cho, Hwasoo Yeo, Nikolas Geroliminis (2025). Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery, Transportation Research Part C: Emerging Technologies, vol. 178, 105205. DOI: 10.1016/j.trc.2025.105205

BibTeX entry:

@article{fonod2025advanced,
 title = {Advanced computer vision for extracting georeferenced vehicle trajectories from drone imagery}, 
 author = {Fonod, Robert and Cho, Haechan and Yeo, Hwasoo and Geroliminis, Nikolas},
 journal = {Transportation Research Part C: Emerging

In this short paper, we study platooning control of drones using only the information from a camera attached to each drone. For this, we adopt real-time objection detection based on a deep learning model called YOLO (you only look once). The YOLO object detector continuously estimates the relative position of the drone in front, by which each drone is controlled by a PD (Proportional-Derivative) feedback controller for platooning. The effectiveness of the proposed system is shown by indoor experiments with three drones.

Overview

This dataset contains 74 images of aerial maritime photographs taken with via a Mavic Air 2 drone and 1,151 bounding boxes, consisting of docks, boats, lifts, jetskis, and cars. This is a multi class problem. This is an aerial object detection dataset. This is a maritime object detection dataset.

The drone was flown at 400 ft. No drones were harmed in the making of this dataset.

This dataset was collected and annotated by the Roboflow team, released with MIT license.

https://i.imgur.com/9ZYLQSO.jpg" alt="Image example">

Use Cases

• Identify number of boats on the water over a lake via quadcopter.
• Boat object detection dataset
• Aerial Object Detection proof of concept
• Identify if boat lifts have been taken out via a drone
• Identify cars with a UAV drone
• Find which lakes are inhabited and to which degree.
• Identify if visitors are visiting the lake house via quad copter.
• Proof of concept for UAV imagery project
• Proof of concept for maritime project
• Etc.

This dataset is a great starter dataset for building an aerial object detection model with your drone.

Getting Started

Fork or download this dataset and follow our How to train state of the art object detector YOLOv4 for more. Stay tuned for particular tutorials on how to teach your UAV drone how to see and comprable airplane imagery and airplane footage.

Annotation Guide

See here for how to use the CVAT annotation tool that was used to create this dataset.

About Roboflow

Roboflow makes managing, preprocessing, augmenting, and versioning datasets for computer vision seamless. :fa-spacer: Developers reduce 50% of their boilerplate code when using Roboflow's workflow, save training time, and increase model reproducibility. :fa-spacer:

The VisDrone-MOT dataset is a large-scale benchmark for multiple object tracking under drone scenes.

The UAVDT dataset is an open-source dataset specifically designed for drone-based detection and tracking. It aims to provide researchers with high-quality and rich multi-task data to facilitate the application of drones in complex environments, particularly for tasks such as object detection, object tracking, and motion analysis.The dataset was released by the Computer Vision Laboratory at Shenzhen University in 2019 and has been widely utilized in areas such as drone video analysis, autonomous driving, and intelligent surveillance.

Enhanced animal welfare has emerged as a pivotal element in contemporary precision animal husbandry, with bovine monitoring constituting a significant facet of precision agriculture. The evolution of intelligent agriculture in recent years has significantly facilitated the integration of drone flight monitoring tools and innovative systems, leveraging deep learning to interpret bovine behavior. Smart drones, outfitted with monitoring systems, have evolved into viable solutions for wildlife protection and monitoring as well as animal husbandry. Nevertheless, challenges arise under actual and multifaceted ranch conditions, where scale alterations, unpredictable movements, and occlusions invariably influence the accurate tracking of unmanned aerial vehicles (UAVs). To address these challenges, this manuscript proposes a tracking algorithm based on deep learning, adhering to the Joint Detection Tracking (JDT) paradigm established by the CenterTrack algorithm. This algorithm is designed to satisfy the requirements of multi-objective tracking in intricate practical scenarios. In comparison with several preeminent tracking algorithms, the proposed Multi-Object Tracking (MOT) algorithm demonstrates superior performance in Multiple Object Tracking Accuracy (MOTA), Multiple Object Tracking Precision (MOTP), and IDF1. Additionally, it exhibits enhanced efficiency in managing Identity Switches (ID), False Positives (FP), and False Negatives (FN). This algorithm proficiently mitigates the inherent challenges of MOT in complex, livestock-dense scenarios.

SeaDronesSee is a large-scale data set aimed at helping develop systems for Search and Rescue (SAR) using Unmanned Aerial Vehicles (UAVs) in maritime scenarios. Building highly complex autonomous UAV/drone systems that aid in SAR missions requires robust computer vision algorithms to detect and track objects or persons of interest. This data set provides three tracks: object detection, single-object tracking, and multi-object tracking.

All datasets and other information can be found at: https://seadronessee.cs.uni-tuebingen.de/home

This dataset contains only the compressed version of the Object Detection v2 Dataset

The UAV Aerial Photography Multi-Object Dataset is designed for smart transportation applications, featuring a collection of internet-collected UAV (Unmanned Aerial Vehicle) aerial photography images with a resolution of 1920 x 1080 pixels. This dataset predominantly covers large-scale scenes such as parking lots and highways, with each image containing over 200 vehicles. Every object within these images is meticulously annotated with a bounding box that aligns with the object's orientation, ensuring precise vehicle detection and tracking.

Enhanced animal welfare has emerged as a pivotal element in contemporary precision animal husbandry, with bovine monitoring constituting a significant facet of precision agriculture. The evolution of intelligent agriculture in recent years has significantly facilitated the integration of drone flight monitoring tools and innovative systems, leveraging deep learning to interpret bovine behavior. Smart drones, outfitted with monitoring systems, have evolved into viable solutions for wildlife protection and monitoring as well as animal husbandry. Nevertheless, challenges arise under actual and multifaceted ranch conditions, where scale alterations, unpredictable movements, and occlusions invariably influence the accurate tracking of unmanned aerial vehicles (UAVs). To address these challenges, this manuscript proposes a tracking algorithm based on deep learning, adhering to the Joint Detection Tracking (JDT) paradigm established by the CenterTrack algorithm. This algorithm is designed to satisfy the requirements of multi-objective tracking in intricate practical scenarios. In comparison with several preeminent tracking algorithms, the proposed Multi-Object Tracking (MOT) algorithm demonstrates superior performance in Multiple Object Tracking Accuracy (MOTA), Multiple Object Tracking Precision (MOTP), and IDF1. Additionally, it exhibits enhanced efficiency in managing Identity Switches (ID), False Positives (FP), and False Negatives (FN). This algorithm proficiently mitigates the inherent challenges of MOT in complex, livestock-dense scenarios.

Seraphim Drone Detection Dataset

  Dataset Overview

This is a comprehensive drone image dataset curated from 23 open-source datasets and processed through a custom cleaning pipeline. The dataset is designed for training object detection models to identify drones in various environments and conditions. The majority of images feature rotary-wing (multi-rotor) unmanned aerial vehicles (UAVs), with a smaller portion representing fixed-wing and hybrid.… See the full description on the dataset page: https://huggingface.co/datasets/lgrzybowski/seraphim-drone-detection-dataset.