Transcript Human Gesture Recognition Using Kinect Camera

Human Gesture Recognition
Using Kinect Camera
Orasa Patsadu, Chakarida Nukoolkit and Bunthit Watanapa
Presented by Carolina Vettorazzo and Diego Santo
1
Introduction

This work proposes a comparison of
human gesture recognition using data
mining classification methods

The gestures where chosen to be the
knowledge base of a smart home system
which monitors and detects the fall
motion of the elderly or hospital patients.
Introduction

Human gesture
◦ Hands, arms, and body
◦ Movements of the head, face, and eyes

Performance of recognition methods
◦
◦
◦
◦
Light conditions
Shadows
Camera angle
Occlusion
The Kinect
The Kinect - depth image

A pattern of IR dots is projected from the
sensor

These dots are detected by the IR camera

The dots will change position based on
how far the objects are from the source.
The Kinect - depth image
The Kinect - depth image
Shotton et al, CVPR(2011)
The Kinect - Skeleton
The Kinect - applications

Kinect Gesture Recognition REALTIME

Kinect-based Hand Gesture Recognition

http://kinectpowerpoint.codeplex.com/
The Kinect - applications

Rehabilitation.

Improvement of athletes performance.

Interactive surfaces.

3D modeling.

Augmented reality
Methodology

Data mining classification
◦ It is the process of extracting valid, previously
unseen or unknown, comprehensible
information from large databases
◦ Algorithms can involve artificial
intelligence, machine learning, statistics,
and database systems.

z-score normalization
◦ improve the accuracy and efficiency of mining
algorithms
Classification Methods
In this study, were selected four popular data mining
classification method were selected :
◦ Back Propagation Neural Network (BPNN)
◦ Support Vector Machine (SVM)
◦ Decision Tree
◦ Naїve Bayes
To identify three human gestures:
◦ Stand
◦ Sit down
◦ Lie down
Classification Methods

Process of Classification
Figure 1: Overview of the proposed system
Classification Methods

Process of Classification
◦ 1,200 input vectors for each of the three
human gesture classes in input data
◦ 3,600 input vectors (x,y,z) for each distance
setting as shown (Stand, Sit down, Lie Down).
◦ 7,200 input vectors in total for both camera
distance settings (2m and 3m)
◦ 1,200 vectors for both camera distance
settings (2m and 3m)
◦ The output data contain 3,600 vectors in total
Classification Methods

Backpropagation Neural Network(BPNN)
◦ BPNN is a multilayer feed forward neural network,
which uses backpropagation algorithm in its learning.
Classification Methods

Support Vector Machine (SVM)
◦ In machine learning, support vector machines (SVMs,
also support vector networks) are supervised learning
models with associated learning algorithms that analyze data and
recognize patterns
Classification Methods

Decision Tree (DT)
◦ Decision Tree is used to classify data from
class label
Classification Methods

Naïve Bayes (NB)
◦ Is a statistical classification which predicts
class membership based on conditional
probabilities.
Human Gestures
Results




BPNN
SVM
DT
NB
100%
99.75%
93.19%
81.94%
Questions
???