Sign Language Recognition

UsingHidden Markov

ModelPresented by:

Vipul Agarwal - 070905060

Outline#INTRODUCTION #SIGN LANGUAGE#PRE-PROCESSING #SKIN AND HAND

DETECTION#OPTICAL FLOW ANALYSIS#FEATURE EXTRACTION FOR

TRAINING DATA#HIDDEN MARKOV MODEL &

ITS USE#PROGRESS REPORT#DEMONSTRATION

Introduction#Interaction with computers may often not

be a comfortable experience.

#Computers should be able to communicate with people with body language.

#Hand gesture recognition becomes important …– Interactive human-machine interface and

virtual environment

Introduction#Two common technologies for hand

gesture recognition

– GLOVE-BASED METHOD• Using special glove-based device to extract

hand posture

– VISION-BASED METHOD• 3D hand/arm modeling• Appearance modeling

Introduction

#3D hand/arm modeling– Highly computational complexity – Using many approximation process

#Appearance modeling– Low computational complexity– Real-time processing

Sign Language#Rely on the hearing society#Two main elements:

– Low and simple level signed alphabet, mimics the letters of the spoken language.

– Higher level signed language, using actions to mimic the meaning or description of the sign.

#The project aim is to make the computer recognize low and simple level American Sign Language.

Sign Language

#American Sign Language

#26 signs to denote the alphabets.

#10 signs to denote numbers

Pre - ProcessingThe video sequence used has a lot of noise due to:

#Low quality of the webcam

#Improper lighting conditions

#Background

Pre - Processing

Pre-processing involves reducing the noise and illumination problems.The morphological operations used for reducing the noise involves:

#Dilation#Statistical Elimination

Pre - ProcessingDILATION>#A disc shaped region is traversed over

every blob and the ones which do not fit the disc are removed completely.

Pre - ProcessingSTATISTICAL ELIMINATION>

#For every region the area is computed. Since hand is the one with the largest area, all blobs having less than a specified area are removed.

Hand Detection#First all the noise is removed in the

pre-processing stage.#Now we assume that the hand is the

largest skin blob in our video sequence.

#We calculate the area of every blob and take the one with the largest area.

#We also calculate the bounding box of the region containing the hand for further analysis

Hand Detection

Optical Flow Analysis

DEFINITION:#Optical flow is the pattern of

apparent motion of objects, surfaces, and edges in a visual scene caused by the relative motion between an observer (an eye or a camera) and the scene.

Optical Flow Analysis

Why Optical Flow Analysis?#Till now the system is just

able to detect the hand and follow the bounding box as the hand moves.

#The problem now is that we need to define a way to take a snapshot of the hand when the hand is not moving.

Optical Flow Analysis

Using this technique we find the motion in the hand. When the hand has stabilized, we assume that the gesture is ready. We then take a snapshot of the hand and perform the recognition on that image.

Feature ExtractionFor training the network with test images we perform the following feature extraction technique:-#Thresholding of the test hand#Converting to a binary image#Finding the centroid of the hand and

orientation of the minor axis.#Making feature vectors using a predefined

number of features.

Feature Extraction

#Extracting the intersection of the feature vectors with the boundary points.

#Finding the scalar length of the vectors from the centroid.

#Normalising the lengths in a scale of 1 to 100 to make it scaling invariant.

Feature Extraction

Hidden Markov Model (HMM)

• HMMs allow you to estimate probabilities of unobserved events

• Given plain text, which underlying parameters generated the surface

HMMs and their Usage• HMMs are very

common in Computational Linguistics:

– GESTURE RECOGNITION (observed: image, hidden: alphabets)

Progress ReportWORK COMPLETED:#Data Collection#Pre-processing #Skin And Hand

Detection#Optical Flow Analysis#Feature Extraction For

Training DataWORK REMAINING:#Training The Hidden

Markov Model

Any Questions …?