public void evaluateObservationNoise(
double minPixelError, double maxPixelError, int N, boolean isPlanar) {
System.out.println("------------------------");
rand = new Random(234);
for (int i = 0; i <= N; i++) {
double mag = (maxPixelError - minPixelError) * i / N + minPixelError;
init(NUM_POINTS, false, isPlanar);
addPixelNoise(mag);
if (!target.process(observationPose, found))
throw new RuntimeException("Not expected to fail");
double expectedEuler[] = RotationMatrixGenerator.matrixToEulerXYZ(motion.getR());
double foundEuler[] = RotationMatrixGenerator.matrixToEulerXYZ(found.getR());
Vector3D_F64 expectedTran = motion.getT();
Vector3D_F64 foundTran = found.getT();
double errorTran = expectedTran.distance(foundTran);
double errorEuler = 0;
double sum = 0;
for (int j = 0; j < 3; j++) {
double e = expectedEuler[j] - foundEuler[j];
errorEuler += e * e;
sum += expectedEuler[j];
}
errorEuler = 100 * Math.sqrt(errorEuler) / Math.sqrt(sum);
System.out.printf("%3d angle %6.2f%% translation %6.2e\n", i, errorEuler, errorTran);
}
}
public void evaluateMinimal(double pixelSigma, boolean isPlanar, int numTrials) {
// make sure each test case has the same random seed
rand = new Random(234);
double totalEuler = 0;
double totalTran = 0;
int totalFail = 0;
for (int i = 0; i < numTrials; i++) {
init(target.getMinimumPoints(), false, isPlanar);
addPixelNoise(pixelSigma);
if (!target.process(observationPose, found)) {
totalFail++;
continue;
}
double expectedEuler[] = RotationMatrixGenerator.matrixToEulerXYZ(motion.getR());
double foundEuler[] = RotationMatrixGenerator.matrixToEulerXYZ(found.getR());
Vector3D_F64 expectedTran = motion.getT();
Vector3D_F64 foundTran = found.getT();
double errorTran = expectedTran.distance(foundTran);
double errorEuler = 0;
double sum = 0;
for (int j = 0; j < 3; j++) {
double e = expectedEuler[j] - foundEuler[j];
errorEuler += e * e;
sum += expectedEuler[j];
}
errorEuler = 100 * Math.sqrt(errorEuler) / Math.sqrt(sum);
// System.out.println(errorEuler);
totalEuler += errorEuler;
totalTran += errorTran;
}
int N = numTrials - totalFail;
System.out.printf(
"%20s N = %d failed %.3f%% euler = %3f%% tran = %5f\n",
name,
target.getMinimumPoints(),
(totalFail / (double) numTrials),
(totalEuler / N),
(totalTran / N));
}
/** Give it a grid and see if it computed a legitimate homography */
@Test
public void basicTest() {
// create a grid an apply an arbitrary transform to it
PlanarCalibrationTarget config = GenericCalibrationGrid.createStandardConfig();
DenseMatrix64F R = RotationMatrixGenerator.eulerXYZ(0.02, -0.05, 0.01, null);
Vector3D_F64 T = new Vector3D_F64(0, 0, -1000);
Se3_F64 motion = new Se3_F64(R, T);
List<Point2D_F64> observations = GenericCalibrationGrid.observations(motion, config);
// compute the homography
Zhang99ComputeTargetHomography alg = new Zhang99ComputeTargetHomography(config);
assertTrue(alg.computeHomography(observations));
DenseMatrix64F H = alg.getHomography();
// test this homography property: x2 = H*x1
List<Point2D_F64> gridPoints = config.points;
for (int i = 0; i < observations.size(); i++) {
Point2D_F64 a = GeometryMath_F64.mult(H, gridPoints.get(i), new Point2D_F64());
double diff = a.distance(observations.get(i));
assertEquals(0, diff, 1e-8);
}
}
/** The two sets of points match perfectly and with a little bit of noise */
@Test
public void test3D() {
for (double noise = 0; noise <= 0.01; noise += 0.01) {
// can only correct small changes
DenseMatrix64F R = RotationMatrixGenerator.eulerXYZ(0.01, -0.002, 0.03, null);
Vector3D_F64 T = new Vector3D_F64(0.02, 0.03, 0.01);
Se3_F64 tran = new Se3_F64(R, T);
List<Point3D_F64> srcPts = UtilPoint3D_F64.random(-10, 10, 30, rand);
List<Point3D_F64> srcOrig = UtilPoint3D_F64.copy(srcPts);
List<Point3D_F64> modelPts = new ArrayList<Point3D_F64>();
for (Point3D_F64 p : srcPts) {
modelPts.add(SePointOps_F64.transform(tran, p, null));
}
// add noise
UtilPoint3D_F64.noiseNormal(modelPts, noise, rand);
PointModel<Point3D_F64> model = new PointModel<Point3D_F64>(modelPts);
StoppingCondition stop = new StoppingCondition(10, 0.1 * noise / srcPts.size() + 1e-8);
IterativeClosestPoint<Se3_F64, Point3D_F64> alg =
new IterativeClosestPoint<Se3_F64, Point3D_F64>(stop, new MotionSe3PointSVD_F64());
alg.setModel(model);
alg.process(srcPts);
Se3_F64 foundTran = alg.getPointsToModel();
checkTransform(srcOrig, modelPts, foundTran, noise * 10 + 1e-8);
}
}
@Override
public void decode(double[] param, CalibratedPoseAndPoint outputModel) {
int paramIndex = 0;
// first decode the transformation
for (int i = 0; i < numViews; i++) {
// don't decode if it is already known
if (knownView[i]) continue;
Se3_F64 se = outputModel.getWorldToCamera(i);
rotation.setParamVector(param[paramIndex++], param[paramIndex++], param[paramIndex++]);
RotationMatrixGenerator.rodriguesToMatrix(rotation, se.getR());
Vector3D_F64 T = se.getT();
T.x = param[paramIndex++];
T.y = param[paramIndex++];
T.z = param[paramIndex++];
}
// now decode the points
for (int i = 0; i < numPoints; i++) {
Point3D_F64 p = outputModel.getPoint(i);
p.x = param[paramIndex++];
p.y = param[paramIndex++];
p.z = param[paramIndex++];
}
}
@Override
public void encode(CalibratedPoseAndPoint model, double[] param) {
int paramIndex = 0;
// first decode the transformation
for (int i = 0; i < numViews; i++) {
// don't encode if it is already known
if (knownView[i]) continue;
Se3_F64 se = model.getWorldToCamera(i);
// force the "rotation matrix" to be an exact rotation matrix
// otherwise Rodrigues will have issues with the noise
if (!svd.decompose(se.getR())) throw new RuntimeException("SVD failed");
DenseMatrix64F U = svd.getU(null, false);
DenseMatrix64F V = svd.getV(null, false);
CommonOps.multTransB(U, V, R);
// extract Rodrigues coordinates
RotationMatrixGenerator.matrixToRodrigues(R, rotation);
param[paramIndex++] = rotation.unitAxisRotation.x * rotation.theta;
param[paramIndex++] = rotation.unitAxisRotation.y * rotation.theta;
param[paramIndex++] = rotation.unitAxisRotation.z * rotation.theta;
Vector3D_F64 T = se.getT();
param[paramIndex++] = T.x;
param[paramIndex++] = T.y;
param[paramIndex++] = T.z;
}
for (int i = 0; i < numPoints; i++) {
Point3D_F64 p = model.getPoint(i);
param[paramIndex++] = p.x;
param[paramIndex++] = p.y;
param[paramIndex++] = p.z;
}
}