Authors

Qingyuan Feng, Xingyu Xian

Background

This project tackles the pervasive issue of slips and falls in urban environments like NYC by drawing inspiration from the remarkable stability strategies of penguins on slippery surfaces. Building on prior research in penguin pose estimation, it employs advanced computer vision and machine learning techniques to capture and analyze complex motion patterns.

Methodology

A robust multi-camera data acquisition system is used to record detailed video footage of penguin locomotion and a precise calibration process is performed to align this footage with ground reaction force measurements to ensure the dataset’s accuracy. Advanced tools such as DeepLabCut are used for pose estimation, while comparative analyses between frameworks like PyTorch and TensorFlow help optimize model performance. Rigorous statistical evaluations—tracking metrics such as training loss, test loss, and RMSE—were used to further refine the approach. The calibrated data forms a foundation for the development of a comprehensive 3D model of penguin gait dynamics to provide actionable insights for enhancing robotic navigation and reducing fall risks among vulnerable urban populations.

Deliverables

• Technical Report

Datasets