Monitoring vulnerable Houbara bustards, birds of both ecological and cultural significance, and detecting intruders that can pose a threat to their nests, is critical for effective conservation of these iconic species. Deep learning-based object detection offers an efficient solution for automating large-scale monitoring, yet its application to Houbara research has been hindered by the lack of comprehensive datasets. To address this gap, we present a new dataset of 24,318 camera-trap images, including 15,070 Houbara bustard images and 9,248 intruder images, all annotated with bounding boxes. Collected between 2011 and 2023 at various times of the day, and using 14 camera models, this dataset provides high diversity and complexity, enabling studies on Houbaras and other bustard species in similar habitats. We benchmarked 10 state-of-the-art object detection models, demonstrating that YOLOv10 outperforms others across evaluation metrics. This dataset represents a significant contribution to wildlife monitoring and conservation, supporting vulnerable Houbara bustard research while offering a foundation for broader applications by providing a valuable resource for wildlife researchers and practitioners.
Ali et al. (Tue,) studied this question.