HGR Presentation

Embed Size (px)

Citation preview

  • 7/31/2019 HGR Presentation

    1/25

    HAND GESTURE RECOGNITIO

    using Mat

    Artificial Intelligence PL57

    Guide :Mr. S Baskar

    Project Members1026 Tarun

    1027 Sunil1030 Atul

  • 7/31/2019 HGR Presentation

    2/25

    Introduction

    Hand Gesture Recognition is very promising arePattern Recognition, it includes both Vision Gesture Recognition as well as Sensor BRecognition.

    2

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    3/25

  • 7/31/2019 HGR Presentation

    4/25

    Domain Explanation

    American Sign Language is a HandGesture Language which is used byDeaf & Dumb people through outthe world. It contains differentgestures for each letter of theEnglish Alphabet.

    4

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    5/25

    Data Collection

    Source of DataInternet

    Images from Camera

    5

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    6/25

    Feed Forward Neural NetworkALGORITHM USED6

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    7/25

    Feed Forward Neural Network7

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    8/25

    Feed Forward Neural Network

    Hand Gesture Recognition

    8

    Initialization

    Output

    Blame

    Adjust Weight

    Initialize Weights randomly initialize bias

  • 7/31/2019 HGR Presentation

    9/25

    Feed Forward Neural Network

    Hand Gesture Recognition

    9

    Initialization

    Output

    Blame

    Adjust Weight

    Calculate Output for the entire netusing Activation Function

    Output=A( w k*Ik + bias )

  • 7/31/2019 HGR Presentation

    10/25

    Feed Forward Neural Network

    Hand Gesture Recognition

    10

    Initialization

    Output

    ErrorAdjust Weight

    Calculate Error for NeuronsSum Square Error

    Mean Square Error

    Calculate Blame for the each neuro

  • 7/31/2019 HGR Presentation

    11/25

    Feed Forward Neural Network

    Hand Gesture Recognition

    11

    Initialization

    Output

    Blame

    Adjust Weight

    Adjust Weights using Back Propagand Gradient Descent

    Wij = W ij + r * E i * A j ( Ij ) * O i

    r Learning RateEj Blame for Neuron JAj Derivative of Neuron Js ActIj Input in previous stepO j Output in previous step

  • 7/31/2019 HGR Presentation

    12/25

    Working of Software12

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    13/25

    Flow Chart

    Hand Gesture Recognition

    13

    Training SetImages &

    Target Values

    PreprocessImages

    Train NeuralNetwork

    Test ImagePreprocess

    ImageMaO

  • 7/31/2019 HGR Presentation

    14/25

    Pre-Processing

    CaptureRGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Images were takenfrom each specifiedfolders using loop

    14

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    15/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Each and everyimage was convertedinto grayscale usingrgb2gray function

    15

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    16/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    ResizeContrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Images of differentsizes were convertedto common size of80x60

    16

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    17/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Darken Background

    17

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    18/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Edge DetectionEdges in X-Direction

    Edges in Y-Direction

    18

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    19/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    OrientingFeatureVector

    Find Edge DirectionsOrientation liesbetween -180 to 180

    degrees

    19

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    20/25

    Pre-Processing

    Capture

    RGB 2 GRAY

    Resize

    Contrast Reset

    X-Y Gradient

    Orienting

    FeatureVector

    Finally Image isconverted in ColumnVector of Degrees

    20

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    21/25

    Training Neural Network

    Following Code was used to train Neural Networ

    net=newff (minmax(x),[30 target_neurons],{'logsig','logsig'},'tranet=train(net,x,t);

    21

    Hand Gesture Recognition

  • 7/31/2019 HGR Presentation

    22/25

  • 7/31/2019 HGR Presentation

    23/25

  • 7/31/2019 HGR Presentation

    24/25

    Testing

    Hand Gesture Recognition

    24

    Testing is simple, Image is taken from Camera anthen pre-processed and given to neural network. the basis of output vector, Target image is shown

    sim (net,test)

  • 7/31/2019 HGR Presentation

    25/25

    THANK YOU25

    Hand Gesture Recognition