Abstract:

We propose a quick autocalibration method towards Kinect sensors to solve the problems such as large error and slow computation speed. Firstly, the corner points of the RGB image are extracted to form the corner point set. Then the point cloud of checkboard making from the RGB image is projected using the 3D  Hough transform detection method to the depth image in order to autoextract the point cloud corners. Then the point corners in the depth image are extracted. Thirdly, the point cloud corners are projected to the depth image using the least squaring plane fitting algorithm. Then the point cloud corners of the depth images are registered to the corner points of the RGB image. Finally, the 3D coordinates of these corner points are made and the pose transform matrix can be calculated. Experimental results show that the proposed algorithm can extract corner points automatically and decrease the average calibration time from 218 ms to 166ms, thus realizing a quick autocalibration of Kinect sensors.

Key words: Kinect;calibration;point cloud registration;pose estimation