using System;
using System.Drawing;
using System.Drawing.Imaging;
using ImageProcessor;
namespace Tracker
{
/// <summary>
/// Summary description for MeanShiftKernel.
/// </summary>
public class MeanShiftKernel : ITracker
{
private const int MEANSHIFT_ITERATIONS = 15;
private Model _targetModel = null;
private Model _candidateModel = null;
private Model _targetLModel = null;
private Model _candidateLModel = null;
int _previousTargetCentreX = 0;
int _previousTargetCentreY = 0;
float _centreOffsetX = 0;
float _centreOffsetY = 0;
int count = 0;
int _numberOfBins = 0;
public MeanShiftKernel(int numberOfBins)
{
_numberOfBins = numberOfBins;
Processor.NumberOfBins = numberOfBins;
}
public Model TargetModel
{
get {return _targetModel;}
}
public Model CandidateModel
{
get {return _candidateModel;}
}
public Model TargetLModel
{
get {return _targetLModel;}
}
public Model CandidateLModel
{
get {return _candidateLModel;}
}
public int Iterations
{
get {return 0;}
}
public Bitmap BackProjectionImage
{
get {return null; }
}
public void CreateTargetModel(Bitmap bmp, Rectangle window)
{
_targetLModel = Processor.CreateHueModelWithEpanichnekovKernel(bmp, window);
//bmp.Clone(window, PixelFormat.Format24bppRgb).Save("Model\\t.jpg",System.Drawing.Imaging.ImageFormat.Jpeg);
//PrintTargetModel(window);
//Find the centre Yo of the target
_previousTargetCentreX = (window.Left + window.Right) / 2;
_previousTargetCentreY = (window.Top + window.Bottom) / 2;
//string t = "Count: " + count + " window: " + window.ToString() + " msX: " + _previousTargetCentreX.ToString() + " msY: " + _previousTargetCentreY.ToString() + "\r\n";
//PrintWindow(t);
}
public virtual Rectangle Track(Bitmap bmp, Rectangle window)
{
//Get window width & height
int width = window.Width;
int height = window.Height;
//Centre of the window
int Xc = (window.Left + window.Right) / 2;
int Yc = (window.Top + window.Bottom) / 2;
//Normalised Space coordinates & kernel distance
float Xn = 0;
float Yn = 0;
//Meanshift vector
float msX = 0;
float msY = 0;
float weight = 0;
float totalWeight = 0;
//Colour value index
int colourValue = 0;
//Centre of meanshift
int targetCentreX = 0;
int targetCentreY = 0;
//Find the correct bin range
int binRange = 256 / _numberOfBins;
//Find the new candidate model
_candidateLModel = Processor.CreateHueModelWithEpanichnekovKernel(bmp, window);
//Crop the image for less processing time
Bitmap croppedBmp = bmp.Clone(new Rectangle(window.Left, window.Top, window.Right-window.Left, window.Bottom-window.Top),PixelFormat.Format24bppRgb);
string t = "Start X: " + _previousTargetCentreX.ToString() + " Y: " + _previousTargetCentreY.ToString() + "\r\n";;
int loopCount = 0;
//Do maximum of 15 reps
for (loopCount=0; loopCount<MEANSHIFT_ITERATIONS; loopCount++)
{
//Find meanshift vector
for (int y=window.Top; y<window.Bottom; y++)
{
for (int x=window.Left; x<window.Right; x++)
{
//Find the normalised x & y
Xn = 2 * (x - Xc) / (float) width;
Yn = 2 * (y - Yc) / (float) height;
if (float.IsNaN(Xn))
Xn = 0;
if (float.IsNaN(Yn))
Yn = 0;
//Get the colour value at this location which serves as an index to
//the bins
colourValue = Processor.GetHueComponent(bmp, x, y);
//Find meanshift vector components
weight = _targetLModel.Bins[colourValue / binRange] / _candidateLModel.Bins[colourValue / binRange];
if (float.IsNaN(weight))
weight = 0;
msX += Xn * weight;
msY += Yn * weight;
totalWeight += weight;
}
}
//Find the meanshift vector
msX = msX / totalWeight;
msY = msY / totalWeight;
//Find the centre in the image space
targetCentreX = Xc + (int)(msX * width) / 2;
targetCentreY = Yc + (int)(msY * height) / 2;
//Check if the centre has not moved, then the target is localised and break
if ((_previousTargetCentreX == targetCentreX) && (_previousTargetCentreY == targetCentreY))
break;
else
{
//Move the window's centre to the meanshift vector
window = GetNewTargetWindow(window, targetCentreX, targetCentreY);
//window.Offset((int) (_previousTargetCentreX - targetCentreX), (int) (_previousTargetCentreY - targetCentreY));
//update the centre value
_previousTargetCentreX = targetCentreX;
_previousTargetCentreY = targetCentreY;
}
}
t += "Iteration Count: " + loopCount.ToString() + "\r\n";
t += "MS X: " + _previousTargetCentreX.ToString() + " Y: " + _previousTargetCentreY.ToString();
PrintWindow(t);
count++;
//return the new window based on the current centre
//window.Offset((int) (_previousTargetCentreX - targetCentreX), (int) (_previousTargetCentreY - targetCentreY));
return GetNewTargetWindow(window, targetCentreX, targetCentreY);;
//return the new window on the candidate target
//return MoveMeanShiftVector(bmp, window);
}
private Rectangle GetNewTargetWindow(Rectangle window, int meanshiftX, int meanshiftY)
{
int windowX = (window.Left + window.Right) / 2;
int windowY = (window.Top + window.Bottom) /2;
//Centre the new window at windowX & windowY
if (meanshiftX > windowX)
meanshiftX = window.Left + (meanshiftX - windowX);
else
meanshiftX = window.Left - (windowX - meanshiftX);
if (meanshiftY > windowY)
meanshiftY = window.Top + (meanshiftY - windowY);
else
meanshiftY = window.Top - (windowY - meanshiftY);
//look out for negative values
meanshiftX = (meanshiftX < 0) ? 0 : meanshiftX;
meanshiftY = (meanshiftY < 0) ? 0 : meanshiftY;
//adjust the size if going out of the window size: 320*240
if (meanshiftX + window.Width >= 320)
meanshiftX = 320 - window.Width;
if (meanshiftY + window.Height >= 240)
meanshiftY = 240 - window.Height;
return new Rectangle(meanshiftX, meanshiftY, window.Width, window.Height);
}
private Rectangle MoveMeanShiftVector(Bitmap bmp, Rectangle window)
{
//Find meanshift vector in X & Y direction
float meanshiftX = 0;
float meanshiftY = 0;
string t = "";
//Iterate to move the meanshift vector closer to the centre. Loop until the
//current pixel moves less than one pixel from the previous pixel in the image space
for (int i=0; i<15; i++)
{
count++;
//Find the new candidate model
_candidateModel = Processor.CreateHueHistogramWithEpanechnikovKernel(bmp, window);
//bmp.Clone(window, PixelFormat.Format24bppRgb).Save("Model\\c" + count.ToString() + ".jpg",System.Drawing.Imaging.ImageFormat.Jpeg);
//PrintTargetModel();
//PrintCandidateModel(window);
//Find meanshift vector
FindMeanShiftVector2(ref meanshiftX, ref meanshiftY);
if ((_previousTargetCentreX == meanshiftX) && (_previousTargetCentreY == meanshiftY))
break;
else
{
_centreOffsetX = meanshiftX - _previousTargetCentreX;
_centreOffsetY = meanshiftY - _previousTargetCentreY;
_previousTargetCentreX = (int)meanshiftX;
_previousTargetCentreY = (int)meanshiftY;
// window = GetNewSearchWindow2(window, _centreOffsetX, _centreOffsetY);
window = GetNewSearchWindow2(window, meanshiftX, meanshiftY);
}
t += "Count: " + count + " window: " + window.ToString() + " msX: " + meanshiftX.ToString() + " msY: " + meanshiftY.ToString() + "\r\n";
PrintWindow(t);
}
return window;
}
private void PrintWindow(string text)
{
if (text != "")
{
System.IO.StreamWriter file = new System.IO.StreamWriter("Model\\window" + count.ToString() + ".txt");
file.WriteLine(text);
file.Close();
}
}
private void PrintCandidateModel(Rectangle window)
{
string text2 = "";
for (int j=window.Top; j<window.Bottom; j++)
{
for (int i=window.Left; i<window.Right; i++)
{
text2 += _candidateModel.GetCoordinateValue(i, j).ToString() + ",";
}
text2 += "\r\n";
}
System.IO.StreamWriter file = new System.IO.StreamWriter("Model\\candidatemodel" + count.ToString() + ".txt");
file.WriteLine(text2);
file.Close();
}
private void PrintTargetModel(Rectangle window)
{
string text2 = "";
float val = 0;
Bitmap bmp = (Bitmap) Image.FromFile("Model\\0.bmp");
for (int j=window.Top; j<window.Bottom; j++)
{
for (int i=window.Left; i<window.Right; i++)
{
val = _targetModel.GetCoordinateValue(i, j);
text2 += val.ToString() + ",";
bmp.SetPixel(i, j, GetColor(val));
}
text2 += "\r\n";
}
bmp.Save("Model\\target.bmp", ImageFormat.Bmp);
System.IO.StreamWriter file = new System.IO.StreamWriter("Model\\targetmodel" + count.ToString() + ".txt");
file.WriteLine(text2);
file.Close();
}
private Color GetColor(float val)
{
return Color.FromArgb((int) val, Color.Gray);
}
private Rectangle GetNewSearchWindow2(Rectangle window, float meanshiftX, float meanshiftY)
{
float windowX = (float) (window.Left + window.Right) / 2;
float windowY = (float) (window.Top + window.Bottom) /2;
meanshiftX = windowX - (meanshiftX * (float)window.Width) / 2;
meanshiftY = windowY - (meanshiftY * (float)window.Height) / 2;
//Centre the new window at windowX & windowY
if (meanshiftX > windowX)
meanshiftX = (float)window.Left + (meanshiftX - windowX);
else
meanshiftX = (float)window.Left - (windowX - meanshiftX);
if (meanshiftY > windowY)
meanshiftY = (float)window.Top + (meanshiftY - windowY);
else
meanshiftY = (float)window.Top - (windowY - meanshiftY);
//look out for negative values
meanshiftX = (meanshiftX < 0) ? 0 : meanshiftX;
meanshiftY = (meanshiftY < 0) ? 0 : meanshiftY;
//adjust the size if going out of the window size: 320*240
if (meanshiftX + window.Width >= 320)
meanshiftX = 320 - window.Width;
if (meanshiftY + window.Height >= 240)
meanshiftY = 240 - window.Height;
return new Rectangle((int) meanshiftX, (int) meanshiftY, window.Width, window.Height);
}
private Rectangle GetNewSearchWindow(Rectangle window, int windowX, int windowY)
{
int winCentreX = (window.Left + window.Right) / 2;
int winCentreY = (window.Top + window.Bottom) / 2;
//Centre the new window at windowX & windowY
if (windowX >= winCentreX)
windowX = window.Left + (windowX - winCentreX);
else
windowX = window.Left - (winCentreX - windowX);
if (windowY >= winCentreY)
windowY = window.Top + (windowY - winCentreY);
else
windowY = window.Top - (winCentreY - windowY);
//look out for negative values
windowX = (windowX < 0) ? 0 : windowX;
windowY = (windowY < 0) ? 0 : windowY;
//adjust the size if going out of the window size: 320*240
if (windowX + window.Width >= 320)
windowX = 320 - window.Width;
if (windowY + window.Height >= 240)
windowY = 240 - window.Height;
return new Rectangle((int) windowX, (int) windowY, window.Width, window.Height);
}
private void MapNormalisedSpaceToImageSpace(Rectangle window, float normalX, float normalY, ref int imageX, ref int imageY)
{
imageX = (int)((normalX * (float)window.Width /2) + (float)(window.Left + window.Right) / 2);
imageY = (int)((normalY * (float)window.Height /2) + (float)(window.Bottom + window.Top) / 2);
// float winCentreX = (window.Left + window.Right) / 2;
// float winCentreY = (window.Top + window.Bottom) / 2;
//
// if (normalX == 0) //On the centre
// imageX = (int) winCentreX;
// else if (normalX < 0) //In left hand side - negative
// imageX = (int) (winCentreX + (winCentreX - window.Left) * normalX);
// else //In right hand side - positive
// imageX = (int) (winCentreX + (window.Right - winCentreX) * normalX);
//
// if (normalY == 0) //On the centre
// imageY = (int) winCentreY;
// else if (normalY < 0) //In bottom section - negative
// imageY = (int) (winCentreY - (window.Bottom - winCentreY) * normalY);
// else //In top section - positive
// imageY = (int) (winCentreY - (winCentreY - window.Top) * normalY);
//
// if (imageX == 0)
// imageX = window.Left;
//
// if (imageY == 0)
// imageY = window.Top;
//
}
private void FindMeanShiftVector2(ref float meanshiftX, ref float meanshiftY)
{
//Find meanshift vector in X & Y direction
float totalWeight = 0;
float msX = 0;
float msY = 0;
Bin candidateBin = null;
Coordinate candidateCoord = null;
float weight = 0;
meanshiftX = 0;
meanshiftY = 0;
float t = 0;
string text = "";
for (int i=0; i<_candidateModel.Bins.Count; i++)
{
candidateBin = _candidateModel.Bins.Item(i);
for (int j=0; j<candidateBin.Coordinates.Count; j++)
{
candidateCoord = candidateBin.Coordinates.Item(j);
//For this pixel of the candidate, get the Bin value for the same pixel
//in the target model. Use normalised coordinates because the image space
//coordinates would keep on changing as the ROI window moves
t = _targetModel.GetBinValueByCoordinate(candidateCoord.NormalisedX, candidateCoord.NormalisedY);
//Find the weight
weight = (float) Math.Sqrt(t / candidateBin.NormalisedValue);
//If there is 0/0, then use 0 otherwise the actual weight
weight = float.IsNaN(weight)? 0 : weight;
weight = float.IsInfinity(weight) ? 0 : weight;
//text += "X: " + candidateCoord.X.ToString() + " Y: " + candidateCoord.Y.ToString() + " W: " + weight.ToString() + "\r\n";
//Get total of X by multiplying the actual pixel location value by the weight
msX += candidateCoord.NormalisedX * weight;
//Get total of Y by multiplying the actual pixel location value by the weight
msY += candidateCoord.NormalisedY * weight;
//Keep the total weight
totalWeight += weight;
}
}
meanshiftX = msX / totalWeight;
meanshiftY = msY / totalWeight;
// text += "MSX: " + meanshiftX.ToString() + " MSY: " + meanshiftY.ToString();
// System.IO.StreamWriter file = new System.IO.StreamWriter("Model\\meanshift" + count.ToString() + ".txt");
// file.WriteLine(text);
// file.Close();
}
private void FindMeanShiftVector(ref float meanshiftX, ref float meanshiftY)
{
//Find meanshift vector in X & Y direction
float totalWeight = 0f;
float msX = 0f;
float msY = 0f;
Bin candidateBin = null;
Bin targetBin = null;
float weight = 0f;
meanshiftX = 0;
meanshiftY = 0;
Coordinate coord = null;
float val = 0;
Bitmap bmp = (Bitmap) Image.FromFile("Model\\0.bmp");
for (int i=0; i<_candidateModel.Bins.Count; i++)
{
candidateBin = _candidateModel.Bins.Item(i);
targetBin = _targetModel.Bins.Item(i);
for (int j=0; j<candidateBin.Coordinates.Count; j++)
{
coord = candidateBin.Coordinates.Item(j);
float t = _targetModel.Bins.GetCoordinateColorValue(coord.X, coord.Y);
weight = t / coord.ColorValue;
//If there is 0/0, then use 0 otherwise the actual weight
weight = float.IsNaN(weight)? 0 : weight;
weight = float.IsInfinity(weight) ? 0 : weight;
bmp.SetPixel(coord.X, coord.Y, GetColor((int)weight));
//Get total of X by multiplying the actual pixel location value by the weight
val = coord.NormalisedX * weight;
msX += val; //coord.NormalisedX * weight;
//Get total of Y by multiplying the actual pixel location value by the weight
msY += coord.NormalisedY * weight;
//Keep the total weight
totalWeight += weight;
}
}
meanshiftX = msX / totalWeight;
meanshiftY = msY / totalWeight;
bmp.Save("Model\\candidate.bmp", ImageFormat.Bmp);
}
public float BhattacharyyaCoefficient
{
get
{
float coeff = 0;
for (int i=0; i<_targetModel.Bins.Count; i++)
{
coeff += (float) Math.Sqrt(_targetModel.Bins.Item(i).NormalisedValue * _candidateModel.Bins.Item(i).NormalisedValue);
}
return coeff;
}
}
}
}
Submit an article or tip