use of org.ddogleg.fitting.modelset.ransac.Ransac in project BoofCV by lessthanoptimal.
the class FactoryDetectLineAlgs method lineRansac.
/**
* Detects line segments inside an image using the {@link DetectLineSegmentsGridRansac} algorithm.
*
* @see DetectLineSegmentsGridRansac
*
* @param regionSize Size of the region considered. Try 40 and tune.
* @param thresholdEdge Threshold for determining which pixels belong to an edge or not. Try 30 and tune.
* @param thresholdAngle Tolerance in angle for allowing two edgels to be paired up, in radians. Try 2.36
* @param connectLines Should lines be connected and optimized.
* @param imageType Type of single band input image.
* @param derivType Image derivative type.
* @return Line segment detector
*/
public static <I extends ImageGray<I>, D extends ImageGray<D>> DetectLineSegmentsGridRansac<I, D> lineRansac(int regionSize, double thresholdEdge, double thresholdAngle, boolean connectLines, Class<I> imageType, Class<D> derivType) {
ImageGradient<I, D> gradient = FactoryDerivative.sobel(imageType, derivType);
ModelManagerLinePolar2D_F32 manager = new ModelManagerLinePolar2D_F32();
GridLineModelDistance distance = new GridLineModelDistance((float) thresholdAngle);
GridLineModelFitter fitter = new GridLineModelFitter((float) thresholdAngle);
ModelMatcher<LinePolar2D_F32, Edgel> matcher = new Ransac<>(123123, manager, fitter, distance, 25, 1);
GridRansacLineDetector<D> alg;
if (derivType == GrayF32.class) {
alg = (GridRansacLineDetector) new ImplGridRansacLineDetector_F32(regionSize, 10, matcher);
} else if (derivType == GrayS16.class) {
alg = (GridRansacLineDetector) new ImplGridRansacLineDetector_S16(regionSize, 10, matcher);
} else {
throw new IllegalArgumentException("Unsupported derivative type");
}
ConnectLinesGrid connect = null;
if (connectLines)
connect = new ConnectLinesGrid(Math.PI * 0.01, 1, 8);
return new DetectLineSegmentsGridRansac<>(alg, connect, gradient, thresholdEdge, imageType, derivType);
}
use of org.ddogleg.fitting.modelset.ransac.Ransac in project BoofCV by lessthanoptimal.
the class FactoryVisualOdometry method stereoDepth.
/**
* Stereo vision based visual odometry algorithm which runs a sparse feature tracker in the left camera and
* estimates the range of tracks once when first detected using disparity between left and right cameras.
*
* @see VisOdomPixelDepthPnP
*
* @param thresholdAdd Add new tracks when less than this number are in the inlier set. Tracker dependent. Set to
* a value ≤ 0 to add features every frame.
* @param thresholdRetire Discard a track if it is not in the inlier set after this many updates. Try 2
* @param sparseDisparity Estimates the 3D location of features
* @param imageType Type of image being processed.
* @return StereoVisualOdometry
*/
public static <T extends ImageGray<T>> StereoVisualOdometry<T> stereoDepth(double inlierPixelTol, int thresholdAdd, int thresholdRetire, int ransacIterations, int refineIterations, boolean doublePass, StereoDisparitySparse<T> sparseDisparity, PointTrackerTwoPass<T> tracker, Class<T> imageType) {
// Range from sparse disparity
StereoSparse3D<T> pixelTo3D = new StereoSparse3D<>(sparseDisparity, imageType);
Estimate1ofPnP estimator = FactoryMultiView.computePnP_1(EnumPNP.P3P_FINSTERWALDER, -1, 2);
final DistanceModelMonoPixels<Se3_F64, Point2D3D> distance = new PnPDistanceReprojectionSq();
ModelManagerSe3_F64 manager = new ModelManagerSe3_F64();
EstimatorToGenerator<Se3_F64, Point2D3D> generator = new EstimatorToGenerator<>(estimator);
// 1/2 a pixel tolerance for RANSAC inliers
double ransacTOL = inlierPixelTol * inlierPixelTol;
ModelMatcher<Se3_F64, Point2D3D> motion = new Ransac<>(2323, manager, generator, distance, ransacIterations, ransacTOL);
RefinePnP refine = null;
if (refineIterations > 0) {
refine = FactoryMultiView.refinePnP(1e-12, refineIterations);
}
VisOdomPixelDepthPnP<T> alg = new VisOdomPixelDepthPnP<>(thresholdAdd, thresholdRetire, doublePass, motion, pixelTo3D, refine, tracker, null, null);
return new WrapVisOdomPixelDepthPnP<>(alg, pixelTo3D, distance, imageType);
}
use of org.ddogleg.fitting.modelset.ransac.Ransac in project BoofCV by lessthanoptimal.
the class ExampleFundamentalMatrix method robustFundamental.
/**
* Given a set of noisy observations, compute the Fundamental matrix while removing
* the noise.
*
* @param matches List of associated features between the two images
* @param inliers List of feature pairs that were determined to not be noise.
* @return The found fundamental matrix.
*/
public static DMatrixRMaj robustFundamental(List<AssociatedPair> matches, List<AssociatedPair> inliers) {
// used to create and copy new instances of the fit model
ModelManager<DMatrixRMaj> managerF = new ModelManagerEpipolarMatrix();
// Select which linear algorithm is to be used. Try playing with the number of remove ambiguity points
Estimate1ofEpipolar estimateF = FactoryMultiView.computeFundamental_1(EnumFundamental.LINEAR_7, 2);
// Wrapper so that this estimator can be used by the robust estimator
GenerateEpipolarMatrix generateF = new GenerateEpipolarMatrix(estimateF);
// How the error is measured
DistanceFromModelResidual<DMatrixRMaj, AssociatedPair> errorMetric = new DistanceFromModelResidual<>(new FundamentalResidualSampson());
// Use RANSAC to estimate the Fundamental matrix
ModelMatcher<DMatrixRMaj, AssociatedPair> robustF = new Ransac<>(123123, managerF, generateF, errorMetric, 6000, 0.1);
// Estimate the fundamental matrix while removing outliers
if (!robustF.process(matches))
throw new IllegalArgumentException("Failed");
// save the set of features that were used to compute the fundamental matrix
inliers.addAll(robustF.getMatchSet());
// Improve the estimate of the fundamental matrix using non-linear optimization
DMatrixRMaj F = new DMatrixRMaj(3, 3);
ModelFitter<DMatrixRMaj, AssociatedPair> refine = FactoryMultiView.refineFundamental(1e-8, 400, EpipolarError.SAMPSON);
if (!refine.fitModel(inliers, robustF.getModelParameters(), F))
throw new IllegalArgumentException("Failed");
// Return the solution
return F;
}
use of org.ddogleg.fitting.modelset.ransac.Ransac in project BoofCV by lessthanoptimal.
the class FactoryMultiViewRobust method epipolarRansac.
private static Ransac<Se3_F64, AssociatedPair> epipolarRansac(Estimate1ofEpipolar epipolar, CameraPinholeRadial intrinsic, ConfigRansac ransac) {
TriangulateTwoViewsCalibrated triangulate = FactoryMultiView.triangulateTwoGeometric();
ModelManager<Se3_F64> manager = new ModelManagerSe3_F64();
ModelGenerator<Se3_F64, AssociatedPair> generateEpipolarMotion = new Se3FromEssentialGenerator(epipolar, triangulate);
DistanceFromModel<Se3_F64, AssociatedPair> distanceSe3 = new DistanceSe3SymmetricSq(triangulate, intrinsic.fx, intrinsic.fy, intrinsic.skew, intrinsic.fx, intrinsic.fy, intrinsic.skew);
double ransacTOL = ransac.inlierThreshold * ransac.inlierThreshold * 2.0;
return new Ransac<>(ransac.randSeed, manager, generateEpipolarMotion, distanceSe3, ransac.maxIterations, ransacTOL);
}
use of org.ddogleg.fitting.modelset.ransac.Ransac in project MAVSlam by ecmnet.
the class FactoryMAVOdometry method depthDepthPnP.
/**
* Depth sensor based visual odometry algorithm which runs a sparse feature tracker in the visual camera and
* estimates the range of tracks once when first detected using the depth sensor.
*
* @see MAVOdomPixelDepthPnP
*
* @param thresholdAdd Add new tracks when less than this number are in the inlier set. Tracker dependent. Set to
* a value ≤ 0 to add features every frame.
* @param thresholdRetire Discard a track if it is not in the inlier set after this many updates. Try 2
* @param sparseDepth Extracts depth of pixels from a depth sensor.
* @param visualType Type of visual image being processed.
* @param depthType Type of depth image being processed.
* @return StereoVisualOdometry
*/
public static <Vis extends ImageGray, Depth extends ImageGray> MAVDepthVisualOdometry<Vis, Depth> depthDepthPnP(double inlierPixelTol, int thresholdAdd, int thresholdRetire, int ransacIterations, int refineIterations, boolean doublePass, DepthSparse3D<Depth> sparseDepth, PointTrackerTwoPass<Vis> tracker, Class<Vis> visualType, Class<Depth> depthType) {
// Range from sparse disparity
ImagePixelTo3D pixelTo3D = new DepthSparse3D_to_PixelTo3D<Depth>(sparseDepth);
Estimate1ofPnP estimator = FactoryMultiView.computePnP_1(EnumPNP.P3P_FINSTERWALDER, -1, 2);
final DistanceModelMonoPixels<Se3_F64, Point2D3D> distance = new PnPDistanceReprojectionSq();
ModelManagerSe3_F64 manager = new ModelManagerSe3_F64();
EstimatorToGenerator<Se3_F64, Point2D3D> generator = new EstimatorToGenerator<Se3_F64, Point2D3D>(estimator);
// 1/2 a pixel tolerance for RANSAC inliers
double ransacTOL = inlierPixelTol * inlierPixelTol;
ModelMatcher<Se3_F64, Point2D3D> motion = new Ransac<Se3_F64, Point2D3D>(2323, manager, generator, distance, ransacIterations, ransacTOL);
RefinePnP refine = null;
if (refineIterations > 0) {
refine = FactoryMultiView.refinePnP(1e-12, refineIterations);
}
MAVOdomPixelDepthPnP<Vis> alg = new MAVOdomPixelDepthPnP<Vis>(thresholdAdd, thresholdRetire, doublePass, motion, pixelTo3D, refine, tracker, null, null);
return new MAVOdomPixelDepthPnP_to_DepthVisualOdometry<Vis, Depth>(sparseDepth, alg, distance, ImageType.single(visualType), depthType);
}
Aggregations