Deep Learning Based Robust Human Body Segmentation for Pose Estimation from RGB-D Sensors
Loading...
Authors
Frank, David Q.
Issue Date
2016
Type
Thesis
Language
Keywords
Convolutional Neural Networks , Deep Learning , Human Computer Interaction , Human Robot Interaction , Pose Estimation , RGB-D Sensor
Alternative Title
Abstract
This project focuses on creating a system for human body segmentation meant to be used for pose estimation. Recognizing a human figure in a cluttered environment is a challenging problem. Current systems for pose estimation assume that there are no objects around the person, which restricts their use in a real world scenario. This project is based on new advances in deep learning, a field of machine learning that can tackle tough vision problems. The system contains a whole pipeline for training and using a system to estimate the pose of a human. It contains a data generation module that creates the training data for the deep learning module. The deep learning module is the main contribution of this work and provides a robust method for segmenting the body parts of a human. Finally, the project includes a pose estimation module which focuses on reducing the detailed output of the deep learning module into a pose skeleton.
Description
Citation
Publisher
License
In Copyright(All Rights Reserved)