/// <summary>
/// 建構子
/// </summary>
/// <param name="indices"></param>
/// <param name="homography">用來繪製ROI的矩陣</param>
/// <param name="mask"></param>
/// <param name="matchedCount">匹配點</param>
/// <param name="template">樣板特徵類別</param>
public SURFMatchedData(Matrix <int> indices, HomographyMatrix homography, Matrix <byte> mask, int matchedCount, SURFFeatureData template)
{
this.indices = indices;
this.homography = homography;
this.mask = mask;
this.matchedCount = matchedCount;
this.templateSURFData = template;
}
/// <summary>
/// 顯示畫出匹配的視窗
/// </summary>
/// <param name="matchData">匹配後回傳的資料類別</param>
/// <param name="observedScene">觀察景象特徵資料</param>
public static void ShowSURFMatchForm(SURFMatchedData matchData, SURFFeatureData observedScene, ImageViewer viewer)
{
//Draw the matched keypoints
Image <Bgr, Byte> result = Features2DToolbox.DrawMatches(matchData.GetTemplateSURFData().GetImg(), matchData.GetTemplateSURFData().GetKeyPoints(), observedScene.GetImg(), observedScene.GetKeyPoints(),
matchData.GetIndices(), new Bgr(255, 255, 255), new Bgr(255, 255, 255), matchData.GetMask(), Features2DToolbox.KeypointDrawType.DEFAULT);
if (matchData.GetHomography() != null) //Get RoI box
{
PointF[] matchPts = GetMatchBoundingBox(matchData.GetHomography(), matchData.GetTemplateSURFData());
if (matchPts != null)
{
result.DrawPolyline(Array.ConvertAll <PointF, Point>(matchPts, Point.Round), true, new Bgr(Color.Red), 2);
}
}
viewer.Image = result;
viewer.Show();
}
/// <summary>
/// 取得對應出物體的ROI座標點
/// </summary>
/// <param name="homography">保存了相關匹配後的資訊(用來投影至商品上的匹配位置)</param>
/// <param name="template">樣板特徵類別</param>
/// <returns>回傳座標點</returns>
public static PointF[] GetMatchBoundingBox(HomographyMatrix homography, SURFFeatureData template)
{
if (homography != null) //Get RoI box
{
//draw a rectangle along the projected model
PointF[] pts = new PointF[] {
new PointF(template.GetImg().ROI.Left, template.GetImg().ROI.Bottom),
new PointF(template.GetImg().ROI.Right, template.GetImg().ROI.Bottom),
new PointF(template.GetImg().ROI.Right, template.GetImg().ROI.Top),
new PointF(template.GetImg().ROI.Left, template.GetImg().ROI.Top)
};
homography.ProjectPoints(pts); //project points
return(pts);
}
else
{
return(null);
}
}
/// <summary>
/// 商品辨識使用BruteForce匹配(較精確但較慢)
/// </summary>
/// <param name="template">樣板的特徵點類別</param>
/// <param name="observedScene">被觀察的場景匹配的特徵點</param>
/// <returns>回傳匹配的資料類別</returns>
public static SURFMatchedData MatchSURFFeatureByBruteForceForGoods(SURFFeatureData template, SURFFeatureData observedScene)
{
//This matrix indicates which row is valid for the matches.
Matrix <byte> mask;
//Number of nearest neighbors to search for
int k = 2;
//The distance different ratio which a match is consider unique, a good number will be 0.8 , NNDR match
double uniquenessThreshold = 0.8; //default:0.8
//The resulting n*k matrix of descriptor index from the training descriptors
Matrix <int> indices;
HomographyMatrix homography = null;
Stopwatch watch;
try
{
watch = Stopwatch.StartNew();
#region bruteForce match for CPU
//match
BruteForceMatcher <float> matcher = new BruteForceMatcher <float>(DistanceType.L2Sqr); //default:L2
matcher.Add(template.GetDescriptors());
indices = new Matrix <int>(observedScene.GetDescriptors().Rows, k);
//The resulting n*k matrix of distance value from the training descriptors
using (Matrix <float> dist = new Matrix <float>(observedScene.GetDescriptors().Rows, k))
{
matcher.KnnMatch(observedScene.GetDescriptors(), indices, dist, k, null);
#region Test Output
//for (int i = 0; i < indices.Rows; i++)
//{
// for (int j = 0; j < indices.Cols; j++)
// {
// Console.Write(indices[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n distance");
//for (int i = 0; i < dist.Rows; i++)
//{
// for (int j = 0; j < dist.Cols; j++)
// {
// Console.Write(dist[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n");
#endregion
mask = new Matrix <byte>(dist.Rows, 1);
mask.SetValue(255); //mask is 拉式信號
//http://stackoverflow.com/questions/21932861/how-does-features2dtoolbox-voteforuniqueness-work
//how the VoteForUniqueness work...
Features2DToolbox.VoteForUniqueness(dist, uniquenessThreshold, mask);
}
int nonZeroCount = CvInvoke.cvCountNonZero(mask); //means good match
Console.WriteLine("-----------------\nVoteForUniqueness pairCount => " + nonZeroCount.ToString() + "\n-----------------");
if (nonZeroCount >= 4)
{
//50 is model and mathing image rotation similarity ex: m1 = 60 m2 = 50 => 60 - 50 <=50 so is similar
nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 1.5, 50); //default:1.5 , 10
Console.WriteLine("VoteForSizeAndOrientation pairCount => " + nonZeroCount.ToString() + "\n-----------------");
if (nonZeroCount >= 15) //defalut :4 ,set 15
{
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 5);
}
}
#endregion
watch.Stop();
Console.WriteLine("Cal SURF Match time => " + watch.ElapsedMilliseconds.ToString() + "\n-----------------");
return(new SURFMatchedData(indices, homography, mask, nonZeroCount, template));
}
catch (CvException ex)
{
System.Windows.Forms.MessageBox.Show(ex.ErrorMessage);
return(null);
}
}
/// <summary>
/// 環境看板辨識使用BruteForce匹配(較精確但較慢)
/// </summary>
/// <param name="template">樣板的特徵點類別</param>
/// <param name="observedScene">被觀察的場景匹配的特徵點</param>
/// <returns>回傳匹配的資料類別</returns>
public static SURFMatchedData MatchSURFFeatureByBruteForceForObjs(SURFFeatureData template, SURFFeatureData observedScene)
{
//This matrix indicates which row is valid for the matches.
Matrix <byte> mask;
//Number of nearest neighbors to search for
int k = 5;
//The distance different ratio which a match is consider unique, a good number will be 0.8 , NNDR match
double uniquenessThreshold = 0.5; //default 0.8
//The resulting n*k matrix of descriptor index from the training descriptors
Matrix <int> trainIdx;
HomographyMatrix homography = null;
Stopwatch watch;
try
{
watch = Stopwatch.StartNew();
#region Surf for CPU
//match
BruteForceMatcher <float> matcher = new BruteForceMatcher <float>(DistanceType.L2Sqr);
matcher.Add(template.GetDescriptors());
trainIdx = new Matrix <int>(observedScene.GetDescriptors().Rows, k);
//The resulting n*k matrix of distance value from the training descriptors
using (Matrix <float> distance = new Matrix <float>(observedScene.GetDescriptors().Rows, k))
{
matcher.KnnMatch(observedScene.GetDescriptors(), trainIdx, distance, k, null);
mask = new Matrix <byte>(distance.Rows, 1);
mask.SetValue(255); //Mask is 拉式信號匹配
//http://stackoverflow.com/questions/21932861/how-does-features2dtoolbox-voteforuniqueness-work
//how the VoteForUniqueness work...
Features2DToolbox.VoteForUniqueness(distance, uniquenessThreshold, mask);
}
Image <Bgr, byte> result = null;
int nonZeroCount = CvInvoke.cvCountNonZero(mask); //means good match
Console.WriteLine("VoteForUniqueness nonZeroCount=> " + nonZeroCount.ToString());
if (nonZeroCount >= (template.GetKeyPoints().Size * 0.2)) //set 10
{
//50 is model and mathing image rotation similarity ex: m1 = 60 m2 = 50 => 60 - 50 <=50 so is similar
nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(template.GetKeyPoints(), observedScene.GetKeyPoints(), trainIdx, mask, 1.2, 50); //default 1.5,10
Console.WriteLine("VoteForSizeAndOrientation nonZeroCount=> " + nonZeroCount.ToString());
if (nonZeroCount >= (template.GetKeyPoints().Size * 0.5)) //default 4 ,set 15
{
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(template.GetKeyPoints(), observedScene.GetKeyPoints(), trainIdx, mask, 5);
}
PointF[] matchPts = GetMatchBoundingBox(homography, template);
//Draw the matched keypoints
result = Features2DToolbox.DrawMatches(template.GetImg(), template.GetKeyPoints(), observedScene.GetImg(), observedScene.GetKeyPoints(),
trainIdx, new Bgr(255, 255, 255), new Bgr(255, 255, 255), mask, Features2DToolbox.KeypointDrawType.NOT_DRAW_SINGLE_POINTS);
if (matchPts != null)
{
result.DrawPolyline(Array.ConvertAll <PointF, Point>(matchPts, Point.Round), true, new Bgr(Color.Red), 2);
}
}
#endregion
watch.Stop();
Console.WriteLine("\nCal SURF Match time=======\n=> " + watch.ElapsedTicks.ToString() + "ms\nCal SURF Match time=======");
return(new SURFMatchedData(trainIdx, homography, mask, nonZeroCount, template));
}
catch (CvException ex)
{
System.Windows.Forms.MessageBox.Show(ex.ErrorMessage);
return(null);
}
}
/// <summary>
/// 匹配較快速但精確度較低
/// </summary>
/// <param name="template">樣板的特徵點類別</param>
/// <param name="observedScene">被觀察的場景匹配的特徵點</param>
/// <returns>回傳匹配的資料類別</returns>
public static SURFMatchedData MatchSURFFeatureByFLANNForObjs(SURFFeatureData template, SURFFeatureData observedScene)
{
Matrix <byte> mask;
int k = 1;
List <int> maskIndex = new List <int>();
double uniquenessThreshold = 0.5;//NNDR
//The resulting n*k matrix of descriptor index from the training descriptors,存放找到的NN索引
Matrix <int> indices;
HomographyMatrix homography = null;
Stopwatch watch;
//distance
Matrix <float> dists;
Index flann;
try
{
watch = Stopwatch.StartNew();
//match
flann = new Index(template.GetDescriptors(), 4);
indices = new Matrix <int>(observedScene.GetDescriptors().Rows, k);
//dists是對應indices索引的距離值
using (dists = new Matrix <float>(observedScene.GetDescriptors().Rows, k))
{
//找出最好的NN
flann.KnnSearch(observedScene.GetDescriptors(), indices, dists, k, 4);
#region Test Output
//for (int i = 0; i < indices.Rows; i++)
//{
// for (int j = 0; j < indices.Cols; j++)
// {
// Console.Write(indices[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n distance");
//for (int i = 0; i < dists.Rows; i++)
//{
// for (int j = 0; j < dists.Cols; j++)
// {
// Console.Write(dists[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n");
#endregion
mask = new Matrix <byte>(dists.Rows, 1);
mask.SetValue(255);
//此emgucv是NNDR,已實驗過,如果要使用VoteForUniqueness 請把mask改回255,mask存放的是樣板與觀察對應的特徵點是否相似0表不是,255表一樣
//Features2DToolbox.VoteForUniqueness(dists, uniquenessThreshold, mask);
//如下,數值會一樣
//for (int i = 0; i < indices.Rows; i++)
//{
// Console.WriteLine("距離比：" + (dists.Data[i, 0] / dists.Data[i, 1]));
// if ((dists.Data[i, 0] / dists.Data[i, 1]) < 0.8)
// {
// mask.Data[i, 0] = 255;
// }
//}
double min_dist = 100;
double max_dist = 0;
//FLANN 取自http://docs.opencv.org/doc/tutorials/features2d/feature_flann_matcher/feature_flann_matcher.html#feature-flann-matcher
//很微妙的是這個方法匹配效果更加...
for (int i = 0; i < indices.Rows; i++)
{
if (dists.Data[i, 0] < min_dist)
{
min_dist = dists.Data[i, 0];
}
if (dists.Data[i, 0] > max_dist)
{
max_dist = dists.Data[i, 0];
}
}
for (int i = 0; i < indices.Rows; i++)
{
if (dists.Data[i, 0] >= Math.Max(2 * min_dist, 0.06))
{
mask.Data[i, 0] = 0;
}
}
//for (int i = 0; i < mask.Rows; i++)
//{
// Console.WriteLine(mask.Data[i, 0]);
//}
}
int nonZeroCount = CvInvoke.cvCountNonZero(mask);
//Console.WriteLine("good Match number:" + nonZeroCount + ",template keypoint number = " + template.GetKeyPoints().Size);
//Console.WriteLine("-----------------\nVoteForUniqueness pairCount => " + nonZeroCount.ToString() + "\n-----------------");
//因為是小圖比大圖，可能有特徵點重複比對到觀察影像導致駔後留下的特徵做於樣板特徵，因為多數情境下，一個畫面中只需要比對出一個即可，應該不需要超過樣板特特徵...
//nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 1.2, 20);
//Console.WriteLine("VoteForSizeAndOrientation pairCount => " + nonZeroCount.ToString() + "\n-----------------");
//filter out all unnecessary pairs based on distance between pairs
for (int i = 0; i < mask.Rows; i++)
{
if (mask.Data[i, 0] == 255)
{
maskIndex.Add(i);
}
}
for (int i = 0; i < maskIndex.Count; i++)
{
for (int j = i + 1; j < maskIndex.Count; j++)
{
double distance_ratio = (Math.Pow(template.GetKeyPoints()[indices[maskIndex[i], 0]].Point.X - template.GetKeyPoints()[indices[maskIndex[j], 0]].Point.X, 2) +
Math.Pow(template.GetKeyPoints()[indices[maskIndex[i], 0]].Point.Y - template.GetKeyPoints()[indices[maskIndex[j], 0]].Point.Y, 2)) /
(Math.Pow(observedScene.GetKeyPoints()[maskIndex[i]].Point.X - observedScene.GetKeyPoints()[maskIndex[j]].Point.X, 2) +
Math.Pow(observedScene.GetKeyPoints()[maskIndex[i]].Point.Y - observedScene.GetKeyPoints()[maskIndex[j]].Point.Y, 2));
//Console.WriteLine(distance_ratio);
if (distance_ratio > 1 || distance_ratio == 0)
{
mask[maskIndex[i], 0] = 0;
}
}
}
//for (int i = 0; i < mask.Rows; i++)
//{
// Console.WriteLine(mask.Data[i, 0]);
//}
nonZeroCount = CvInvoke.cvCountNonZero(mask);
watch.Stop();
//Console.WriteLine("mask=" + nonZeroCount + ",Cal SURF Match time => " + watch.ElapsedMilliseconds.ToString() + "\n-----------------");
if (template.GetKeyPoints().Size > nonZeroCount && nonZeroCount >= template.GetKeyPoints().Size * 0.5)
{
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 5); //原先是5
PointF[] matchPts = GetMatchBoundingBox(homography, template);
if (matchPts != null)
{
return(new SURFMatchedData(indices, homography, mask, nonZeroCount, template));
}
return(null);
}
return(null);
}
catch (CvException ex)
{
System.Windows.Forms.MessageBox.Show(ex.ErrorMessage);
return(null);
}
}
/// <summary>
/// 匹配較快速但精確度較低
/// </summary>
/// <param name="template">樣板的特徵點類別</param>
/// <param name="observedScene">被觀察的場景匹配的特徵點</param>
/// <returns>回傳匹配的資料類別</returns>
public static SURFMatchedData MatchSURFFeatureByFLANNForGoods(SURFFeatureData template, SURFFeatureData observedScene)
{
Matrix <byte> mask;
int k = 1;
double uniquenessThreshold = 0.8;//NNDR
//The resulting n*k matrix of descriptor index from the training descriptors,存放找到的NN索引
Matrix <int> indices;
HomographyMatrix homography = null;
Stopwatch watch;
//distance
Matrix <float> dists;
try
{
watch = Stopwatch.StartNew();
#region FLANN Match CPU
//match
Index flann = new Index(template.GetDescriptors(), 12);
indices = new Matrix <int>(observedScene.GetDescriptors().Rows, k);
//dists是對應indices索引的距離值
using (dists = new Matrix <float>(observedScene.GetDescriptors().Rows, k))
{
//找出最好的NN
flann.KnnSearch(observedScene.GetDescriptors(), indices, dists, k, 12);
#region Test Output
//for (int i = 0; i < indices.Rows; i++)
//{
// for (int j = 0; j < indices.Cols; j++)
// {
// Console.Write(indices[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n distance");
//for (int i = 0; i < dists.Rows; i++)
//{
// for (int j = 0; j < dists.Cols; j++)
// {
// Console.Write(dists[i, j] + " ");
// }
// Console.Write("\n");
//}
//Console.WriteLine("\n");
#endregion
mask = new Matrix <byte>(dists.Rows, 1);
mask.SetValue(0);
//此emgucv是NNDR,已實驗過,如果要使用VoteForUniqueness 請把mask改回255,mask存放的是樣板與觀察對應的特徵點是否相似0表不是,255表一樣
// Features2DToolbox.VoteForUniqueness(dists, uniquenessThreshold, mask);
//如下,數值會一樣
//for (int i = 0; i < indices.Rows; i++)
//{
// Console.WriteLine("距離比：" + (dists.Data[i, 0] / dists.Data[i, 1]));
// if ((dists.Data[i, 0] / dists.Data[i, 1]) < 0.8)
// {
// mask.Data[i, 0] = 255;
// }
//}
double min_dist = 100;
double max_dist = 0;
//FLANN 取自http://docs.opencv.org/doc/tutorials/features2d/feature_flann_matcher/feature_flann_matcher.html#feature-flann-matcher
//很微妙的是這個方法匹配效果更加...
for (int i = 0; i < indices.Rows; i++)
{
if (dists.Data[i, 0] < min_dist)
{
min_dist = dists.Data[i, 0];
}
if (dists.Data[i, 0] > max_dist)
{
max_dist = dists.Data[i, 0];
}
}
for (int i = 0; i < indices.Rows; i++)
{
if (dists.Data[i, 0] <= Math.Max(2 * min_dist, 0.02))
{
mask.Data[i, 0] = 255;
}
}
//for (int i = 0; i < mask.Rows; i++)
//{
// Console.WriteLine(mask.Data[i, 0]);
//}
}
int nonZeroCount = CvInvoke.cvCountNonZero(mask);
Console.WriteLine("good Match number:" + nonZeroCount);
//Console.WriteLine("-----------------\nVoteForUniqueness pairCount => " + nonZeroCount.ToString() + "\n-----------------");
if (nonZeroCount >= 4) //原先是4
{
//nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 1.2, 50);
//Console.WriteLine("VoteForSizeAndOrientation pairCount => " + nonZeroCount.ToString() + "\n-----------------");
//filter out all unnecessary pairs based on distance between pairs
//if (nonZeroCount >= 30) //原先是4
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(template.GetKeyPoints(), observedScene.GetKeyPoints(), indices, mask, 5); //原先是5
}
#endregion
watch.Stop();
Console.WriteLine("Cal SURF Match time => " + watch.ElapsedMilliseconds.ToString() + "\n-----------------");
return(new SURFMatchedData(indices, homography, mask, nonZeroCount, template));
}
catch (CvException ex)
{
System.Windows.Forms.MessageBox.Show(ex.ErrorMessage);
return(null);
}
}
