Animal identification using machine learning techniques
Presentation en seminarioiii
1. Background
Problem Definition
Methodology
Extraction and Tracking of a body Skeleton
from Multiple views
Master’s thesis proposal
Alexander Pinzon Fernandez
August 31, 2009
Alexander Pinzon Fernandez Bioingenium Research Group
2. Background
Problem Definition
Methodology
Outline
1 Background
Background
2 Problem Definition
Problem
Objectives
3 Methodology
Methodology
Alexander Pinzon Fernandez Bioingenium Research Group
3. Background
Problem Definition Background
Methodology
Outline
1 Background
Background
2 Problem Definition
Problem
Objectives
3 Methodology
Methodology
Alexander Pinzon Fernandez Bioingenium Research Group
4. Background
Problem Definition Background
Methodology
Background
The study of the human body has been of interest in several areas.
For example anatomy, engineering and arts.
The movement record has been used to solve different problems.
Diagnosis of gait disorders.
Motion capture for computer character animation [3].
Advanced ergonomics analysis and design.
Alexander Pinzon Fernandez Bioingenium Research Group
5. Background
Problem
Problem Definition
Objectives
Methodology
Outline
1 Background
Background
2 Problem Definition
Problem
Objectives
3 Methodology
Methodology
Alexander Pinzon Fernandez Bioingenium Research Group
6. Background
Problem
Problem Definition
Objectives
Methodology
Problem Definition
Most traditional motion tracking methods are based on optical
systems and present the following disadvantages:
• The use of markers attached to the body ALTER the movement
gesture.
• The need of experts to place the markers because they must be
located in specific anthropometric points.
Alexander Pinzon Fernandez Bioingenium Research Group
7. Background
Problem
Problem Definition
Objectives
Methodology
Problem Definition
The stereo systems that perform a three-dimensional
reconstruction MUST handle large volumes of data of the body
geometry. This process requires high-performance machines
Cost between 60,000 and 130,000 dollars (Data for 2009)
Qualisys AB , Sweden. 8 Oqus cameras: 100.000 USD
Sports Motion, Inc . USA, California. 8 DV Cameras: 60.000
USD
BTS Spa, Italy. 10 infrared cameras Smart-DDigital: 132.000
USD
Alexander Pinzon Fernandez Bioingenium Research Group
8. Background
Problem
Problem Definition
Objectives
Methodology
Description Project
To develop a method to extract the skeleton of an articulated body,
under the following conditions:
• No body model.
• Not having a model is an advantage because it is possible to
extract and track the articulated skeleton of any body.
• The extracted skeleton is a synthesized representation of the
body geometry.
• Once the skeleton is extracted, it is followed for each frame of the
video
Alexander Pinzon Fernandez Bioingenium Research Group
9. Background
Problem
Problem Definition
Objectives
Methodology
Outline
1 Background
Background
2 Problem Definition
Problem
Objectives
3 Methodology
Methodology
Alexander Pinzon Fernandez Bioingenium Research Group
10. Background
Problem
Problem Definition
Objectives
Methodology
Objectives
Overall Objective: To develop a method for motion tracking of the
human body in 3D.
Specific Objectives:
1. To implement a video-camera based system to capture a body
motion, and implement the Σ − ∆ Sigma-Delta method to extract
the silhouette of the body from videos.
2. To propose a method for extracting body markers which
correspond to the fundamental body relations.
3. To propose a method to track the skeleton in each video-frame
and validate the results.
4. To develop a system to visualize the body movement together
with the estimated skeleton.
Alexander Pinzon Fernandez Bioingenium Research Group
11. Background
Problem Definition Methodology
Methodology
Outline
1 Background
Background
2 Problem Definition
Problem
Objectives
3 Methodology
Methodology
Alexander Pinzon Fernandez Bioingenium Research Group
12. Background
Problem Definition Methodology
Methodology
Methodology
Skeleton Extraction and Motion Tracking
• Data acquisition.
• Silhouette Segmentation.
• 3D Reconstruction from the data.
• Skeleton extraction of the reconstructed three-dimensional object.
Alexander Pinzon Fernandez Bioingenium Research Group
13. Background
Problem Definition Methodology
Methodology
Data Acquisition and Segmentation
Objective: To implement a video-cameras based system to capture a body motion
Video Capture system (named optical system) is divided into two
problems, Camera Calibration and Synchronized Multicamera
[1].
The image segmentation of this videos is realized with a set of
methods for dividing an image into regions, given certain
characteristics[2].
Alexander Pinzon Fernandez Bioingenium Research Group
14. Background
Problem Definition Methodology
Methodology
Three Dimensional Reconstruction
Objective: To propose a method for extracting markers of the body fundamental relations
using the videos.
3D Reconstruction steps:
Shape from silhouette
Visuall Hull
Find characteristic points
Extract characteristic points using visual attention model
Stereoscopic reconstruction
Surface reconstruction with the photometric method called
stereo pair, using shape and characteristic points
Alexander Pinzon Fernandez Bioingenium Research Group
15. Background
Problem Definition Methodology
Methodology
Skeleton Extraction
Skeleton extraction is the process of synthesizing and representing a
body in a 1D structure
Based on medial axis concept.
Validation:
• Centered skeleton is its centeredness within the object
• Homotopic skeleton have the same number of connected
components, tunnels, and cavities.
Alexander Pinzon Fernandez Bioingenium Research Group
16. Background
Problem Definition Methodology
Methodology
Bibliography
G.K.M. Cheung, S. Baker, and T. Kanade.
Visual hull alignment and refinement across time: a 3d reconstruction algorithm
combining shape-from-silhouette with stereo.
In Computer Vision and Pattern Recognition, 2003. Proceedings. 2003 IEEE
Computer Society Conference on, volume 2, pages II–375–82 vol.2, June 2003.
Ning Jin and F. Mokhtarian.
Image-based shape model for view-invariant human motion recognition.
In Advanced Video and Signal Based Surveillance, pages 336–341, Sept. 2007.
Fabio Remondino.
3-d reconstruction of static human body shape from image sequence.
Computer Vision and Image Understanding, 93:65–85, 2004.
Alexander Pinzon Fernandez Bioingenium Research Group