Stanford Drone Dataset: Multi-scale, Multi-target social navigation

Stanford researchers have created the first large-scale dataset of aerial videos from multiple classes of targets interacting in complex outdoor spaces. The Stanford Drone Dataset comprises of more than 100 different top-view scenes for a total of 20,000 targets engaged in various types of interactions. Previous datasets included mostly single target but this dataset includes pedestrians, bikers, skaters, cars, and carts, all interacting based on common sense and social etiquette rules. Moreover, the researchers introduced a new characterization that describes the "social sensitivity" at which two targets interact. This feature is used to define "navigation styles" which improve both forecasting models and state-of-the-art multi-target tracking such as those used in computer vision models.

Stanford Drone Dataset

Stage of Research

• Proof-of-concept
• Demonstrated the ability to model the interaction between humans and their surroundings to improve or solve numerous computer vision tasks