public Mat Describe(Mat image, out KeyPoint[] keypoints)
{
Mat descriptors = new Mat();
_featureDetector.DetectAndCompute(image, null, out keypoints, descriptors);
return(descriptors);
}
public StopSignDetector(IInputArray stopSignModel)
{
_detector = new KAZE();
using (Mat redMask = new Mat())
{
GetRedPixelMask(stopSignModel, redMask);
_modelKeypoints = new VectorOfKeyPoint();
_modelDescriptors = new Mat();
_detector.DetectAndCompute(redMask, null, _modelKeypoints, _modelDescriptors, false);
if (_modelKeypoints.Size == 0)
{
throw new Exception("No image feature has been found in the stop sign model");
}
}
_modelDescriptorMatcher = new BFMatcher(DistanceType.L2);
_modelDescriptorMatcher.Add(_modelDescriptors);
_octagon = new VectorOfPoint(
new Point[]
{
new Point(1, 0),
new Point(2, 0),
new Point(3, 1),
new Point(3, 2),
new Point(2, 3),
new Point(1, 3),
new Point(0, 2),
new Point(0, 1)
});
}
private void FindStopSign(Mat img, List <Mat> stopSignList, List <Rectangle> boxList, VectorOfVectorOfPoint contours, int[,] hierachy, int idx)
{
for (; idx >= 0; idx = hierachy[idx, 0])
{
using (VectorOfPoint c = contours[idx])
using (VectorOfPoint approx = new VectorOfPoint())
{
CvInvoke.ApproxPolyDP(c, approx, CvInvoke.ArcLength(c, true) * 0.02, true);
double area = CvInvoke.ContourArea(approx);
if (area > 200)
{
double ratio = CvInvoke.MatchShapes(_octagon, approx, Emgu.CV.CvEnum.ContoursMatchType.I3);
if (ratio > 0.1) //not a good match of contour shape
{
//check children
if (hierachy[idx, 2] >= 0)
{
FindStopSign(img, stopSignList, boxList, contours, hierachy, hierachy[idx, 2]);
}
continue;
}
Rectangle box = CvInvoke.BoundingRectangle(c);
Mat candidate = new Mat();
using (Mat tmp = new Mat(img, box))
CvInvoke.CvtColor(tmp, candidate, ColorConversion.Bgr2Gray);
//set the value of pixels not in the contour region to zero
using (Mat mask = new Mat(candidate.Size.Height, candidate.Width, DepthType.Cv8U, 1))
{
mask.SetTo(new MCvScalar(0));
CvInvoke.DrawContours(mask, contours, idx, new MCvScalar(255), -1, LineType.EightConnected, null, int.MaxValue, new Point(-box.X, -box.Y));
double mean = CvInvoke.Mean(candidate, mask).V0;
CvInvoke.Threshold(candidate, candidate, mean, 255, ThresholdType.Binary);
CvInvoke.BitwiseNot(candidate, candidate);
CvInvoke.BitwiseNot(mask, mask);
candidate.SetTo(new MCvScalar(0), mask);
}
int minMatchCount = 8;
double uniquenessThreshold = 0.8;
VectorOfKeyPoint _observeredKeypoint = new VectorOfKeyPoint();
Mat _observeredDescriptor = new Mat();
_detector.DetectAndCompute(candidate, null, _observeredKeypoint, _observeredDescriptor, false);
if (_observeredKeypoint.Size >= minMatchCount)
{
int k = 2;
Mat mask;
using (VectorOfVectorOfDMatch matches = new VectorOfVectorOfDMatch())
{
_modelDescriptorMatcher.KnnMatch(_observeredDescriptor, matches, k, null);
mask = new Mat(matches.Size, 1, DepthType.Cv8U, 1);
mask.SetTo(new MCvScalar(255));
Features2DToolbox.VoteForUniqueness(matches, uniquenessThreshold, mask);
}
int nonZeroCount = CvInvoke.CountNonZero(mask);
if (nonZeroCount >= minMatchCount)
{
boxList.Add(box);
stopSignList.Add(candidate);
}
}
}
}
}
}
public static void FindMatch(
Mat modelImage,
Mat observedImage,
Feature2D featureDetectorExtractor,
out long matchTime,
out VectorOfKeyPoint modelKeyPoints,
out VectorOfKeyPoint observedKeyPoints,
VectorOfVectorOfDMatch matches,
out Mat mask,
out Mat homography)
{
int k = 2;
double uniquenessThreshold = 0.80;
Stopwatch watch;
homography = null;
modelKeyPoints = new VectorOfKeyPoint();
observedKeyPoints = new VectorOfKeyPoint();
using (UMat uModelImage = modelImage.GetUMat(AccessType.Read))
using (UMat uObservedImage = observedImage.GetUMat(AccessType.Read))
{
//extract features from the object image
Mat modelDescriptors = new Mat();
featureDetectorExtractor.DetectAndCompute(uModelImage, null, modelKeyPoints, modelDescriptors, false);
watch = Stopwatch.StartNew();
// extract features from the observed image
Mat observedDescriptors = new Mat();
featureDetectorExtractor.DetectAndCompute(uObservedImage, null, observedKeyPoints, observedDescriptors, false);
// Brute force, slower but more accurate
// You can use KDTree for faster matching with slight loss in accuracy
using (Emgu.CV.Flann.LinearIndexParams ip = new Emgu.CV.Flann.LinearIndexParams())
using (Emgu.CV.Flann.SearchParams sp = new SearchParams())
using (DescriptorMatcher matcher = new FlannBasedMatcher(ip, sp))
{
matcher.Add(modelDescriptors);
matcher.KnnMatch(observedDescriptors, matches, k, null);
mask = new Mat(matches.Size, 1, DepthType.Cv8U, 1);
mask.SetTo(new MCvScalar(255));
Features2DToolbox.VoteForUniqueness(matches, uniquenessThreshold, mask);
int nonZeroCount = CvInvoke.CountNonZero(mask);
if (nonZeroCount >= 4)
{
nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(modelKeyPoints, observedKeyPoints,
matches, mask, 1.5, 20);
if (nonZeroCount >= 4)
{
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(modelKeyPoints,
observedKeyPoints, matches, mask, 2);
}
}
}
watch.Stop();
}
matchTime = watch.ElapsedMilliseconds;
}
public static bool TestFeature2DTracker(Feature2D keyPointDetector, Feature2D descriptorGenerator)
{
//for (int k = 0; k < 1; k++)
{
Feature2D feature2D = null;
if (keyPointDetector == descriptorGenerator)
{
feature2D = keyPointDetector as Feature2D;
}
Mat modelImage = EmguAssert.LoadMat("box.png");
//Image<Gray, Byte> modelImage = new Image<Gray, byte>("stop.jpg");
//modelImage = modelImage.Resize(400, 400, true);
//modelImage._EqualizeHist();
#region extract features from the object image
Stopwatch stopwatch = Stopwatch.StartNew();
VectorOfKeyPoint modelKeypoints = new VectorOfKeyPoint();
Mat modelDescriptors = new Mat();
if (feature2D != null)
{
feature2D.DetectAndCompute(modelImage, null, modelKeypoints, modelDescriptors, false);
}
else
{
keyPointDetector.DetectRaw(modelImage, modelKeypoints);
descriptorGenerator.Compute(modelImage, modelKeypoints, modelDescriptors);
}
stopwatch.Stop();
EmguAssert.WriteLine(String.Format("Time to extract feature from model: {0} milli-sec", stopwatch.ElapsedMilliseconds));
#endregion
//Image<Gray, Byte> observedImage = new Image<Gray, byte>("traffic.jpg");
Image <Gray, Byte> observedImage = EmguAssert.LoadImage <Gray, byte>("box_in_scene.png");
//Image<Gray, Byte> observedImage = modelImage.Rotate(45, new Gray(0.0));
//image = image.Resize(400, 400, true);
//observedImage._EqualizeHist();
#region extract features from the observed image
stopwatch.Reset();
stopwatch.Start();
VectorOfKeyPoint observedKeypoints = new VectorOfKeyPoint();
using (Mat observedDescriptors = new Mat())
{
if (feature2D != null)
{
feature2D.DetectAndCompute(observedImage, null, observedKeypoints, observedDescriptors, false);
}
else
{
keyPointDetector.DetectRaw(observedImage, observedKeypoints);
descriptorGenerator.Compute(observedImage, observedKeypoints, observedDescriptors);
}
stopwatch.Stop();
EmguAssert.WriteLine(String.Format("Time to extract feature from image: {0} milli-sec", stopwatch.ElapsedMilliseconds));
#endregion
//Merge the object image and the observed image into one big image for display
Image <Gray, Byte> res = modelImage.ToImage <Gray, Byte>().ConcateVertical(observedImage);
Rectangle rect = new Rectangle(Point.Empty, modelImage.Size);
PointF[] pts = new PointF[] {
new PointF(rect.Left, rect.Bottom),
new PointF(rect.Right, rect.Bottom),
new PointF(rect.Right, rect.Top),
new PointF(rect.Left, rect.Top)
};
Mat homography = null;
stopwatch.Reset();
stopwatch.Start();
int k = 2;
DistanceType dt = modelDescriptors.Depth == CvEnum.DepthType.Cv8U ? DistanceType.Hamming : DistanceType.L2;
//using (Matrix<int> indices = new Matrix<int>(observedDescriptors.Rows, k))
//using (Matrix<float> dist = new Matrix<float>(observedDescriptors.Rows, k))
using (VectorOfVectorOfDMatch matches = new VectorOfVectorOfDMatch())
using (BFMatcher matcher = new BFMatcher(dt))
{
//ParamDef[] parameterDefs = matcher.GetParams();
matcher.Add(modelDescriptors);
matcher.KnnMatch(observedDescriptors, matches, k, null);
Mat mask = new Mat(matches.Size, 1, DepthType.Cv8U, 1);
mask.SetTo(new MCvScalar(255));
//mask.SetValue(255);
Features2DToolbox.VoteForUniqueness(matches, 0.8, mask);
int nonZeroCount = CvInvoke.CountNonZero(mask);
if (nonZeroCount >= 4)
{
nonZeroCount = Features2DToolbox.VoteForSizeAndOrientation(modelKeypoints, observedKeypoints, matches, mask, 1.5, 20);
if (nonZeroCount >= 4)
{
homography = Features2DToolbox.GetHomographyMatrixFromMatchedFeatures(modelKeypoints, observedKeypoints, matches, mask, 2);
}
}
}
stopwatch.Stop();
EmguAssert.WriteLine(String.Format("Time for feature matching: {0} milli-sec", stopwatch.ElapsedMilliseconds));
bool success = false;
if (homography != null)
{
PointF[] points = pts.Clone() as PointF[];
points = CvInvoke.PerspectiveTransform(points, homography);
//homography.ProjectPoints(points);
for (int i = 0; i < points.Length; i++)
{
points[i].Y += modelImage.Height;
}
res.DrawPolyline(
#if NETFX_CORE
Extensions.
#else
Array.
#endif
ConvertAll <PointF, Point>(points, Point.Round), true, new Gray(255.0), 5);
success = true;
}
//Emgu.CV.UI.ImageViewer.Show(res);
return(success);
}
/*
* stopwatch.Reset(); stopwatch.Start();
* //set the initial region to be the whole image
* using (Image<Gray, Single> priorMask = new Image<Gray, float>(observedImage.Size))
* {
* priorMask.SetValue(1.0);
* homography = tracker.CamShiftTrack(
* observedFeatures,
* (RectangleF)observedImage.ROI,
* priorMask);
* }
* Trace.WriteLine(String.Format("Time for feature tracking: {0} milli-sec", stopwatch.ElapsedMilliseconds));
*
* if (homography != null) //set the initial tracking window to be the whole image
* {
* PointF[] points = pts.Clone() as PointF[];
* homography.ProjectPoints(points);
*
* for (int i = 0; i < points.Length; i++)
* points[i].Y += modelImage.Height;
* res.DrawPolyline(Array.ConvertAll<PointF, Point>(points, Point.Round), true, new Gray(255.0), 5);
* return true;
* }
* else
* {
* return false;
* }*/
}
}
private static bool FindDescriptors(Feature2D feature2D, VectorOfKeyPoint modelKeyPoints, UMat uModelImage, out UMat modelDescriptors)
{
modelDescriptors = new UMat();
feature2D.DetectAndCompute(uModelImage, null, modelKeyPoints, modelDescriptors, false);
return(modelDescriptors.Size.Height + modelDescriptors.Size.Width > 1);
}