Dissertation Talk: Learning to Predict Human Behavior from Video
Seminar | May 8 | 12-1 p.m. | 337A Cory Hall
In recent times, the field of computer vision has made great progress with recognizing and tracking people and their activities in videos. However, for systems designed to interact dynamically with humans, tracking and recognition are insufficient; the ability to predict behavior is requisite. In this talk, I will present my work on learning to make predictions from visual input. Using team sports videos as a case study, I will show generic frameworks for predicting future actions and motion trajectories in multi-agent, adversarial environments. In unconstrained, single-agent environments, I will present a framework for learning a representation of 3D human dynamics that can be used to: a) recover smooth 3D meshes of humans moving in videos, and b) hallucinate the 3D past and future motion from a single input image.