ETIS, ENSEA, UCP, CNRS
I am engineer ENSEEIHT (Institut National Polytechnique Toulouse), Doctor from LAAS Lab in Robotics and Computer Vision and after I teach and research with ENSEA (Cergy-Pontoise) and ETIS respectively
Is speaking at
Thursday, March 21, 2019
Lecture : Stereovision from the 2 cameras of the new HTC vive pro headsetStereovision permits to rebuild in 3D the environment from this device in order to perform mixed reality. The previous version of HTC vive with one camera could not carried out stereovision. This allows us to add computer vision to virtual reality without adding supplementary cameras on the headset. In this work, the work volume is first defined for obtaining stereo images with small geometrical deformations although they can be modelized but this costs an additional CPU time. At the beginning, the 2 cameras are calibrated for estimating the geometrical transform between the scene and the image plane. The calibration grid is dihedral and composed of horizontal and vertical lines in order to get high accuracy of intersection points on the grid and on the image. The scene coordinate system is easily attached to this calibration object while the usual coordinate system is put on the image. The calibration step consists in estimating the 12 parameters in the 2 homographic transformations between the coordinate (x,y,z) of a 3D object point in the scene and its image coordinates (u,v). This parameter estimation is computed from N known points in the scene and their image pixels. The calibration parameters have been checked as well as the perspective transform between the scene and the image. After the geometrical calibration of these 2 cameras, the stereo-matching has been made by using the epipolar geometry obtained from the fundamental matrix evaluated from couples of homologous pixels in the 2 images which are almost rectified. Then, the stereo-matching process is launched. All the intersection points of this dihedral grid are rebuilt in 3D by stereovision and the geometrical precision of these 3D points is estimated utilizing different statistical procedures with the 3D coordinates points measured on the calibration grid. The 2 camera calibrations and stereovision processes have been studied according to the time in order to evaluate the temporal stability of the 2 cameras of this headset. The stereo-vision has also been tested with success on other objects.
- Michel CHAPRON (ETIS, ENSEA, UCP, CNRS)