PT Unknown
AU Wenjuan Gong
TI 3D Motion Data aided Human Action Recognition and Pose Estimation
PY 2013
AB In this work, we explore human action recognition and pose estimation prob-lems. Different from traditional works of learning from 2D images or videosequences and their annotated output, we seek to solve the problems with ad-ditional 3D motion capture information, which helps to fill the gap between 2Dimage features and human interpretations.We first compare two different schools of approaches commonly used for 3Dpose estimation from 2D pose configuration: modeling and learning methods.By looking into experiments results and considering our problems, we fixed alearning method as the following approaches to do pose estimation. We thenestablish a framework by adding a module of detecting 2D pose configurationfrom images with varied background, which widely extend the application ofthe approach. We also seek to directly estimate 3D poses from image features,instead of estimating 2D poses as a intermediate module. We explore a robustinput feature, which combined with the proposed distance measure, providesa solution for noisy or corrupted inputs. We further utilize the above methodto estimate weak poses,which is a concise representation of the original posesby using dimension deduction technologies, from image features. Weak posespace is where we calculate vocabulary and label action types using a bog ofwords pipeline. Temporal information of an action is taken into consideration byconsidering several consecutive frames as a single unit for computing vocabularyand histogram assignments.
ER