Search for a command to run...
The UAV-Gait50 dataset is a specialized collection of aerial walking sequences designed to bridge the gap between laboratory-controlled environments and unpredictable real-world surveillance scenarios. The dataset contains video-captured silhouettes of 50 individuals, recorded using a drone at varying altitudes. This collection is specifically curated to challenge and improve the robustness of gait recognition algorithms against viewpoint variations and clothing-induced occlusions. The data was captured in an outdoor environment using an Unmanned Aerial Vehicle (UAV). To simulate authentic surveillance conditions and prevent altitude-related bias, the drone’s height was dynamically adjusted between 3 meters and 10 meters. This multi-altitude approach provides a rich variety of scale changes and perspective shifts, reflecting the technical challenges faced by aerial monitoring systems. The dataset includes 50 participants, consisting of 40 males and 10 females. A significant feature of this dataset is the emphasis on diverse, real-life clothing: Males: Dressed in standard shirts, T-shirts, and trousers. Females: Dressed in traditional South Asian attire, including Salwar Kameez, Burqas and Hijabs. Intentional Challenge: The inclusion of loose-fitting clothing (like Burqas) intentionally results in non-ideal or "unclear" silhouettes. This simulates the practical difficulty of biometric identification in regions where traditional attire is common, forcing models to learn more robust features beyond simple limb contours. Key Features: Variable Altitude: Captured at heights of 3m to 10m to ensure viewpoint invariance. Real-World Attire: Specifically includes complex silhouettes caused by traditional clothing and Burqas. Natural Movement: Subjects were recorded walking in a natural outdoor setting rather than a restricted indoor track. Surveillance Realism: Designed to test the limits of person identification in real-life security and monitoring applications. Research Significance: Standard gait datasets often rely on clear, side-view silhouettes with tight-fitting clothing. UAV-Gait50 provides a much-needed benchmark for the computer vision community to develop algorithms that can handle: 1. Vertical Scale Variation: Understanding how altitude changes affect gait signatures. 2. Clothing Occlusion: Identifying individuals even when the full limb movement is obscured by loose garments. 3. Aerial Perspective: Addressing the "top-down" distortion inherent in drone-based security.