Example Stereo Disparity 3D

From BoofCV
Jump to: navigation, search

An additional example for stereo disparity. This one shows you how to properly resize input images and convert the disparity image into a 3D point cloud.

Example Code:


  • Stereo disparity
  • Point clouds

Relevant Applets:

Related Examples:

Example Code

 * Expanding upon ExampleStereoDisparity, this example demonstrates how to rescale an image for stereo processing and
 * then compute its 3D point cloud.  Images are often rescaled to improve speed and some times quality.  Creating
 * 3D point clouds from disparity images is easy and well documented in the literature, but there are some nuances
 * to it.
 * @author Peter Abeles
public class ExampleStereoDisparity3D {

	// Specifies what size input images are scaled to
	public static final double scale = 0.5;

	// Specifies what range of disparity is considered
	public static final int minDisparity = 0;
	public static final int maxDisparity = 40;
	public static final int rangeDisparity = maxDisparity-minDisparity;

	public static void main( String args[] ) {
		// ------------- Compute Stereo Correspondence

		// Load camera images and stereo camera parameters
		String calibDir = UtilIO.pathExample("calibration/stereo/Bumblebee2_Chess/");
		String imageDir = UtilIO.pathExample("stereo/");

		StereoParameters param = CalibrationIO.load(new File(calibDir , "stereo.yaml"));

		// load and convert images into a BoofCV format
		BufferedImage origLeft = UtilImageIO.loadImage(imageDir , "chair01_left.jpg");
		BufferedImage origRight = UtilImageIO.loadImage(imageDir , "chair01_right.jpg");

		GrayU8 distLeft = ConvertBufferedImage.convertFrom(origLeft, (GrayU8) null);
		GrayU8 distRight = ConvertBufferedImage.convertFrom(origRight,(GrayU8)null);

		// re-scale input images
		GrayU8 scaledLeft = new GrayU8((int)(distLeft.width*scale),(int)(distLeft.height*scale));
		GrayU8 scaledRight = new GrayU8((int)(distRight.width*scale),(int)(distRight.height*scale));

		new FDistort(distLeft,scaledLeft).scaleExt().apply();
		new FDistort(distRight,scaledRight).scaleExt().apply();

		// Don't forget to adjust camera parameters for the change in scale!
		PerspectiveOps.scaleIntrinsic(param.left, scale);

		// rectify images and compute disparity
		GrayU8 rectLeft = new GrayU8(scaledLeft.width,scaledLeft.height);
		GrayU8 rectRight = new GrayU8(scaledRight.width,scaledRight.height);

		RectifyCalibrated rectAlg = ExampleStereoDisparity.rectify(scaledLeft,scaledRight,param,rectLeft,rectRight);

//		GrayU8 disparity = ExampleStereoDisparity.denseDisparity(rectLeft, rectRight, 3,minDisparity, maxDisparity);
		GrayF32 disparity = ExampleStereoDisparity.denseDisparitySubpixel(rectLeft, rectRight, 3, minDisparity, maxDisparity);

		// ------------- Convert disparity image into a 3D point cloud

		// The point cloud will be in the left cameras reference frame
		DMatrixRMaj rectK = rectAlg.getCalibrationMatrix();
		DMatrixRMaj rectR = rectAlg.getRectifiedRotation();

		// extract intrinsic parameters from rectified camera
		double baseline = param.getBaseline()*0.1;
		double fx = rectK.get(0,0);
		double fy = rectK.get(1,1);
		double cx = rectK.get(0,2);
		double cy = rectK.get(1,2);

		double maxZ = baseline*100;

		// Iterate through each pixel in disparity image and compute its 3D coordinate
		PointCloudViewer pcv = VisualizeData.createPointCloudViewer();

		List<Point3D_F64> temp = new ArrayList<>();
		Point3D_F64 pointRect = new Point3D_F64();
		Point3D_F64 pointLeft = new Point3D_F64();
		for( int y = 0; y < disparity.height; y++ ) {
			for( int x = 0; x < disparity.width; x++ ) {
				double d = disparity.unsafe_get(x,y) + minDisparity;

				// skip over pixels were no correspondence was found
				if( d >= rangeDisparity || d <= 0 )

				// Coordinate in rectified camera frame
				pointRect.z = baseline*fx/d;
				pointRect.x = pointRect.z*(x - cx)/fx;
				pointRect.y = pointRect.z*(y - cy)/fy;

				// prune points which are likely to be noise
				if( pointRect.z >= maxZ )

				// rotate into the original left camera frame
				GeometryMath_F64.multTran(rectR, pointRect, pointLeft);

				// add pixel to the view for display purposes and sets its gray scale value
				int v = rectLeft.unsafe_get(x, y);
				pcv.addPoint(pointLeft.x,pointLeft.y,pointLeft.z,v << 16 | v << 8 | v);
//				temp.add( pointLeft.copy() );

		// move it back a bit to make the 3D structure more apparent
		Se3_F64 cameraToWorld = new Se3_F64();
		cameraToWorld.T.z = -baseline*5;
		cameraToWorld.T.x = baseline*12;

		// Configure the display
//		pcv.addCloud(temp);
//		pcv.setShowAxis(true);
		JComponent viewer = pcv.getComponent();
		viewer.setPreferredSize(new Dimension(600,600*param.left.height/param.left.width));

		// display the results.  Click and drag to change point cloud camera
		BufferedImage visualized = VisualizeImageData.disparity(disparity, null,minDisparity, maxDisparity,0);
		ShowImages.showWindow(visualized,"Disparity", true);
		ShowImages.showWindow(viewer,"Point Cloud", true);