HuBot Dataset: Annotated Data for Non-Disruptive Bird Behavior Study

This dataset was created as part of the HuBot project, which focuses on developing a biomimetic mobile robot for non-invasive ecological research and wildlife monitoring. It contains a curated collection of 5,000 high-quality images of Houbara bustards, annotated for object detection tasks. The dataset aims to support research in ecological conservation and robotic applications, enabling accurate and reliable detection of Houbara bustards in diverse environments.

Dataset Details:
Size: 5,000 images, divided into:
Training set: 4,000 images.
Validation set: 500 images.
Testing set: 500 images.
Annotations: Each image includes bounding box annotations for detecting and localizing Houbara bustards.
Sources: Images were collected from in-house footage, online repositories, and globally deployed camera traps.
Features:
Bias Mitigation:

Includes diverse lighting conditions, backgrounds, and challenging scenarios to avoid overfitting.
Balanced representation of Houbara behaviors and habitats to ensure unbiased performance.
Data Augmentation:

Applied techniques such as horizontal flips, cropping, blurring, and noise addition to enhance model generalization.
Diversity and Representation:

The dataset captures a range of environmental conditions to support real-world applications and improve model robustness.
Future Scaling:

Plans are in place to expand the dataset to 50,000 images, covering rare Houbara behaviors, multiple subspecies, and more geographical locations.
Applications:
Object detection in biomimetic robotics.
Ecological research on endangered species.
Non-invasive data collection for long-term wildlife monitoring.
This dataset has been instrumental in optimizing HuBot’s object detection algorithms, allowing reliable performance in diverse conditions while minimizing disturbance to wildlife. It serves as a valuable resource for researchers and developers working on ecological robotics, conservation, and artificial intelligence.

License: This dataset is shared under the CC BY 4.0 license, allowing use for non-commercial purposes with appropriate attribution.