A Multi-layer Composite Model for Human Pose Estimation

Abstract

We introduce a new approach for part-based human pose estimation using multi-layer composite models, in which each layer is a tree-structured pictorial structure that models pose at a different scale and with a different graphical structure. At the highest level, the submodel acts as a person detector, while at the lowest level, the body is decomposed into a collection of many local parts. Edges between adjacent layers of the composite model encode cross-model constraints. This multi-layer composite model is able to relax the independence assumptions of traditional tree-structured pictorial-structure models while permitting efficient inference using dual-decomposition. We propose an optimization procedure for joint learning of the entire composite model. Our approach outperforms the state-of-the-art on the challenging Parse and UIUC Sport datasets.

Privacy behaviors of lifeloggers using wearable cameras

“…Automated mechanisms could attempt to extract how many people are in the photo, whether certain people are in the photo, whether their expressions are embarrassing, and the activity occurring in the photo. Existing work in face recognition [26] and pose detection [11] could potentially be applied here to produce automated analysis algorithms.…”

Poselet Conditioned Pictorial Structures

“…The most prominent improvement is observed for the torso, but the improvement for upper/lower legs is also pronounced. Our method is slightly better than the multi-layer composite model of [8]. Their approach aims to capture non-tree dependencies between the parts by decomposing the model into multiple layers and performing dual decomposition to cope with cycles in the part graph.…”

PlaceAvoider: Steering First-Person Cameras away from Sensitive Spaces

“…While this general scene categorization would be desirable, computer vision work has shown that recognizing specific targets is much more accurate than recognizing categories of objects; e.g., it is much easier to build a specific model of your bathroom than a general model to recognize any bathroom. Another possibility is to analyze the poses and activities of people in the scene to provide additional evidence to the classifier, using work on people and pose recognition [11], [14]; photos showing people in distress, in compromising poses, or wearing little clothing could be flagged as sensitive. We leave such an exploration to future work.…”