Fast Human Pose Estimation using Appearance and Motion via Multi-Dimensional Boosting Regression UCLA CSD-TR 050046

We address the problem of estimating human pose in video sequences, where the rough location of the human has been detected. We exploit both appearance and motion information by defining suitable features of an image and its temporal neighbors, and learning a regression map to the parameters of a model of the human body using boosting techniques. Our work is intended to bridge the gap between efficient human body detectors that can estimate rough location but not pose in quasi-real time, and computationally expensive but accurate pose estimation algorithms based on dynamic programming. Our algorithm can be viewed as a fast initialization step for human body trackers, or as a tracker itself. In order to accomplish our task, we extend gradient boosting techniques to learn a multi-dimensional map from (rotated and scaled) Haar features to the entire set of joint angles representing the full body pose. Compared to prior work that advocated learning a separate regressor for each joint angle, our approach is more efficient (all joint angle estimators share the same features) and more robust (it exploits the high degree of correlation between joint angles for natural human pose).