A Pedestrian-Vehicle Interaction Benchmark and Annotation Framework for Unstructured Scenes via Uncalibrated Cameras

Researchers introduced PINNS, a public benchmark and scalable annotation framework for dense pedestrian-vehicle interactions in unstructured scenes.

The dataset is built from uncalibrated surveillance videos collected across multiple regions and conditions, with trajectory and scene-level labels.

PINNS aims to fill a major gap for training and evaluating prediction models in complex real-world mixed traffic.

The resource could support safer autonomous driving and better analysis of pedestrian-vehicle behavior in challenging environments.