Example Stereo Uncalibrated

From BoofCV
Jump to: navigation, search
Left is disparity image and Right is found 3D Point Cloud

Given two views from an unknown camera compute a dense 3D point cloud. This is a very challenging problem and many of the steps are inherently unstable. This example code only converges to a valid solution about 40% of the time when presented with two well textured input images with good geometry! When it does work it can work almost well as the calibrated cases! If three views are available the problem becomes much more stable.

See code comments below for a summary of all the processing steps involved.

Example File:


  • Self calibration / Automatic Calibration
  • Projective Geometry
  • Metric Geometry
  • Point feature association
  • Rectification
  • Dense stereo processing

Related Tutorials/Example Code:

Example Code

 * <p>
 * In this example a stereo point cloud is computed from two uncalibrated stereo images. The uncalibrated problem
 * is much more difficult from the calibrated or semi-calibrated problem and the solution below will often fail.
 * The root cause of this difficulty is that it is impossible to remove all false associations given two views,
 * even if the true fundamental matrix is provided! For that reason it is recommended that you use a minimum of
 * three views with uncalibrated observations.
 * </p>
 * <p>
 *     A summary of the algorithm is provided below. There are many ways in which it can be improved upon, but would
 *     increase the complexity. There is also no agreed upon best solution found in the literature and the solution
 *     presented below is "impractical" because of its sensitivity to tuning parameters. If you got a solution
 *     which does a better job let us know!
 * </p>
 * <ol>
 *     <li>Feature association</li>
 *     <li>RANSAC to estimate Fundamental matrix</li>
 *     <li>Guess and check focal length and compute projective to metric homography</li>
 *     <li>Upgrade to metric geometry</li>
 *     <li>Set up bundle adjustment and triangulate 3D points</li>
 *     <li>Run bundle adjustment</li>
 *     <li>Prune outlier points</li>
 *     <li>Run bundle adjustment again</li>
 *     <li>Compute stereo rectification</li>
 *     <li>Rectify images</li>
 *     <li>Compute stereo disparity</li>
 *     <li>Compute and display point cloud</li>
 * </ol>
 * @author Peter Abeles
public class ExampleStereoUncalibrated {

	public static void main( String args[] ) {

		// Successful
		String name = "mono_wall_";
//		String name = "bobcats_";
//		String name = "minecraft_cave1_";
//		String name = "chicken_";
//		String name = "books_";

		// Successful Failures
//		String name = "triflowers_";

		// Failures
//		String name = "rock_leaves_";
//		String name = "minecraft_distant_";
//		String name = "rockview_";
//		String name = "pebbles_";
//		String name = "skull_";
//		String name = "turkey_";

		BufferedImage buff01 = UtilImageIO.loadImage(UtilIO.pathExample("triple/"+name+"01.jpg"));
		BufferedImage buff02 = UtilImageIO.loadImage(UtilIO.pathExample("triple/"+name+"02.jpg"));

		Planar<GrayU8> color01 = ConvertBufferedImage.convertFrom(buff01,true, ImageType.pl(3,GrayU8.class));
		Planar<GrayU8> color02 = ConvertBufferedImage.convertFrom(buff02,true, ImageType.pl(3,GrayU8.class));

		GrayU8 image01 = ConvertImage.average(color01,null);
		GrayU8 image02 = ConvertImage.average(color02,null);

		// Find a set of point feature matches
		List<AssociatedPair> matches = ExampleFundamentalMatrix.computeMatches(buff01, buff02);

		// Prune matches using the epipolar constraint. use a low threshold to prune more false matches
		List<AssociatedPair> inliers = new ArrayList<>();
		DMatrixRMaj F = ExampleFundamentalMatrix.robustFundamental(matches, inliers, 0.1);

		// Perform self calibration using the projective view extracted from F
		// Note that P1 = [I|0]
		System.out.println("Self calibration");
		DMatrixRMaj P2 = MultiViewOps.fundamentalToProjective(F);

		// Take a crude guess at the intrinsic parameters. Bundle adjustment will fix this later.
		int width = buff01.getWidth(), height = buff02.getHeight();
		double fx = width/2; double fy = fx;
		double cx = width/2; double cy = height/2;

		// Compute a transform from projective to metric by assuming we know the camera's calibration
		EstimatePlaneAtInfinityGivenK estimateV = new EstimatePlaneAtInfinityGivenK();

		Vector3D_F64 v = new Vector3D_F64(); // plane at infinity
		if( !estimateV.estimatePlaneAtInfinity(P2,v))
			throw new RuntimeException("Failed!");

		DMatrixRMaj K = PerspectiveOps.pinholeToMatrix(fx,fy,0,cx,cy);
		DMatrixRMaj H = MultiViewOps.createProjectiveToMetric(K,v.x,v.y,v.z,1,null);
		DMatrixRMaj P2m = new DMatrixRMaj(3,4);

		// Decompose and get the initial estimate for translation
		DMatrixRMaj tmp = new DMatrixRMaj(3,3);
		Se3_F64 view1_to_view2 = new Se3_F64();

		//------------------------- Setting up bundle adjustment
		// bundle adjustment will provide a more refined and accurate estimate of these parameters
		System.out.println("Configuring bundle adjustment");

		// Construct bundle adjustment data structure
		SceneStructureMetric structure = new SceneStructureMetric(false);
		SceneObservations observations = new SceneObservations(2);

		// We will assume that the camera has fixed intrinsic parameters
		BundlePinholeSimplified bp = new BundlePinholeSimplified();
		bp.f = fx;

		// The first view is the world coordinate system
		structure.setView(0,true,new Se3_F64());
		// Second view was estimated previously

		for (int i = 0; i < inliers.size(); i++) {
			AssociatedPair t = inliers.get(i);

			// substract out the camera center from points. This allows a simple camera model to be used and
			// errors in the this coordinate tend to be non-fatal

			// each point is visible in both of the views

		// initial location of points is found through triangulation

		//------------------ Running Bundle Adjustment
		System.out.println("Performing bundle adjustment");
		ConfigLevenbergMarquardt configLM = new ConfigLevenbergMarquardt();
		configLM.dampeningInitial = 1e-3;
		configLM.hessianScaling = false;
		ConfigBundleAdjustment configSBA = new ConfigBundleAdjustment();
		configSBA.configOptimizer = configLM;

		// Create and configure the bundle adjustment solver
		BundleAdjustment<SceneStructureMetric> bundleAdjustment = FactoryMultiView.bundleAdjustmentMetric(configSBA);
		// prints out useful debugging information that lets you know how well it's converging
		// Specifies convergence criteria
		bundleAdjustment.configure(1e-6, 1e-6, 100);

		// Scaling improve accuracy of numerical calculations
		ScaleSceneStructure bundleScale = new ScaleSceneStructure();


		// Sometimes pruning outliers help improve the solution. In the stereo case the errors are likely
		// to already fatal
		PruneStructureFromSceneMetric pruner = new PruneStructureFromSceneMetric(structure,observations);


		for (int i = 0; i < structure.cameras.length; i++) {
		for (int i = 0; i < structure.views.length; i++) {

		// display the inlier matches found using the robust estimator
		System.out.println("\n\nComputing Stereo Disparity");
		BundlePinholeSimplified cp = structure.getCameras()[0].getModel();
		CameraPinholeRadial intrinsic = new CameraPinholeRadial();

		Se3_F64 leftToRight = structure.views[1].worldToView;