BaseAbsoluteOrientationSlamTwoViewsSparseReconstructor.java
/*
* Copyright (C) 2017 Alberto Irurueta Carro (alberto@irurueta.com)
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.irurueta.ar.sfm;
import com.irurueta.ar.slam.AbsoluteOrientationBaseSlamEstimator;
import com.irurueta.ar.slam.BaseCalibrationData;
import com.irurueta.geometry.MetricTransformation3D;
import com.irurueta.geometry.Point3D;
import com.irurueta.geometry.Rotation3D;
import java.util.ArrayList;
/**
* Base class in charge of estimating cameras and 3D reconstructed points from
* sparse image point correspondences in two views and also in charge of
* estimating overall scene scale and absolute orientation by means of SLAM
* (Simultaneous Location And Mapping) using data obtained from sensors like
* accelerometers or gyroscopes.
* NOTE: absolute orientation slam estimators are not very accurate during
* estimation of the orientation state, for that reason we take into account
* the initial orientation.
*
* @param <D> type of calibration data.
* @param <C> type of configuration.
* @param <R> type of re-constructor.
* @param <L> type of listener.
* @param <S> type of SLAM estimator.
*/
public abstract class BaseAbsoluteOrientationSlamTwoViewsSparseReconstructor<
D extends BaseCalibrationData,
C extends BaseSlamTwoViewsSparseReconstructorConfiguration<D, C>,
R extends BaseSlamTwoViewsSparseReconstructor<D, C, R, L, S>,
L extends BaseSlamTwoViewsSparseReconstructorListener<R>,
S extends AbsoluteOrientationBaseSlamEstimator<D>> extends
BaseSlamTwoViewsSparseReconstructor<D, C, R, L, S> {
/**
* First sample of orientation received.
*/
protected Rotation3D firstOrientation;
/**
* Constructor.
*
* @param configuration configuration for this re-constructor.
* @param listener listener in charge of handling events.
* @throws NullPointerException if listener or configuration is not
* provided.
*/
protected BaseAbsoluteOrientationSlamTwoViewsSparseReconstructor(
final C configuration, final L listener) {
super(configuration, listener);
}
/**
* Provides a new orientation sample to update SLAM estimator.
* If re-constructor is not running, calling this method has no effect.
*
* @param timestamp timestamp of accelerometer sample since epoch time and
* expressed in nanoseconds.
* @param orientation new orientation.
*/
public void updateOrientationSample(final long timestamp,
final Rotation3D orientation) {
if (slamEstimator != null) {
slamEstimator.updateOrientationSample(timestamp, orientation);
}
if (firstOrientation == null) {
//make a copy of orientation
firstOrientation = orientation.toQuaternion();
}
}
/**
* Updates scene scale and orientation using SLAM data.
*
* @return true if scale was successfully updated, false otherwise.
*/
@SuppressWarnings("DuplicatedCode")
protected boolean updateScaleAndOrientation() {
// obtain baseline (camera separation from slam estimator data
final var posX = slamEstimator.getStatePositionX();
final var posY = slamEstimator.getStatePositionY();
final var posZ = slamEstimator.getStatePositionZ();
// to estimate baseline, we assume that first camera is placed at
// world origin
final var baseline = Math.sqrt(posX * posX + posY * posY + posZ * posZ);
try {
final var camera1 = estimatedCamera1.getCamera();
final var camera2 = estimatedCamera2.getCamera();
camera1.decompose();
camera2.decompose();
final var center1 = camera1.getCameraCenter();
final var center2 = camera2.getCameraCenter();
// R1' = R1*Rdiff
// Rdiff = R1^T*R1'
// where R1' is the desired orientation (obtained by sampling a
// sensor)
// and R1 is always the identity for the 1st camera.
// Hence R1' = Rdiff
// t1' is the desired translation which is zero for the 1st
// camera.
// We want: P1' = K*[R1' t1'] = K*[R1' 0]
// And we have P1 = K[I 0]
// We need a transformation T so that:
// P1' = P1*T^-1 = K[I 0][R1' 0]
// [0 1]
// Hence: T^-1 = [R1' 0]
// [0 1]
// or T = [R1'^T 0]
// [0 1]
// because we are also applying a transformation of scale s,
// the combination of both transformations is
// T = [s*R1'^T 0]
// [0 1]
final var r = firstOrientation.inverseRotationAndReturnNew();
final var estimatedBaseline = center1.distanceTo(center2);
final var scale = baseline / estimatedBaseline;
final var scaleAndOrientationTransformation = new MetricTransformation3D(scale);
scaleAndOrientationTransformation.setRotation(r);
// update scale of cameras
scaleAndOrientationTransformation.transform(camera1);
scaleAndOrientationTransformation.transform(camera2);
estimatedCamera1.setCamera(camera1);
estimatedCamera2.setCamera(camera2);
// update scale of reconstructed points
final var numPoints = reconstructedPoints.size();
final var reconstructedPoints3D = new ArrayList<Point3D>();
for (final var reconstructedPoint : reconstructedPoints) {
reconstructedPoints3D.add(reconstructedPoint.getPoint());
}
scaleAndOrientationTransformation.transformAndOverwritePoints(reconstructedPoints3D);
// set scaled points into result
for (var i = 0; i < numPoints; i++) {
reconstructedPoints.get(i).setPoint(reconstructedPoints3D.get(i));
}
return true;
} catch (final Exception e) {
failed = true;
//noinspection unchecked
listener.onFail((R) this);
return false;
}
}
}