Predicting Pedestrian Behavior from 3D Cues


At a glance

Predicting pedestrian behavior is critical for self driving. Especially when pedestrians decide to jaywalk. Subtle cues of where the human is looking (gaze) and pose gestures are key. For example, people who are about to jay walk tend to look around as they cross. We will explore state-of-the-art human pose and shape estimation models on large-scale driving datasets.

principal investigatorsresearchthemes

Angjoo Kanazawa

Trevor Darrell


action prediction, motion models, human action

This project has continued under the new name: "Spatio-temporal 3D reconstruction of Pedestrians, Objects, and the Environment from Self-Driving Data".