Top left: ocean surface; only one of the videos from a stereo rig is shown. Top right: variational ocean surface reconstruction without calibration refinement. Bottom left: joint ocean surface reconstruction and camera calibration refinement. Bottom right: difference between wave heights with and without calibration refinement.
Top left: synthetic ocean surface generated using CUDA SDK; Top right: simulated perturbations of the camera's extrinsic parameters (rotation and translation). Bottom left: joint ocean surface reconstruction and camera calibration refinement. Bottom right: ocean surface reconstruction without camera calibration refinement.
Space-time reconstruction of Ocean Waves using Variational Stereo Vision (Ph.D. Thesis)
Reconstructed wave height from input stereo images
Left: Images from one of the two calibrated cameras and region of interest to be reconstructed. Right: 3-D reconstructed surface shape of the ocean waves, pseudocolored from blue (low height) to red (large height).
Input stereo images, acquired at a frame rate of 10Hz, are of size 1624 x 1236 pixels.
Parameters of the region of interest. Physical area covered: 13 x 13 m^2. Grid size: 513 x 513 points. Grid resolution: 2.5 cm.
Recovered ocean surface height and "color"
Surface texture (left) and shape (right) given by the variational 3-D reconstruction method of Gallego et al. for the robust 3-D stereoscopic reconstruction of ocean waves from multi-view stereo videos.
Right: surface shape (height) is greyscale encoded: dark means low wave height (troughs) and bright means large wave height (crests).
Modeled stereo images of the ocean waves
Modeled images (within the region of interest) superimposed on the original images. Modeled images are generated using the surface shape (height) and radiance (texture) computed by the variational graph method for the robust 3-D stereoscopic reconstruction of ocean waves from multi-view stereo videos.
Modeled stereo images (showing the region of interest)
Modeled images (within the region of interest) superimposed on the original images. Modeled images are generated using the surface shape (height) and radiance (texture) computed by the variational method of Gallego et al. for the robust 3-D stereoscopic reconstruction of ocean waves from multi-view stereo videos. The reconstructed region has been clearly marked by modifying the intensities in the outside.
Error images: image intensity difference
Error images: absolute difference between original and modeled images within the region of interest, magnified by a factor of 4 to notice it.