Search code examples
computer-visiondatasetkittiodometry

Visual Odometry, Kitti Dataset

I am currently trying to make a stereo visual odometry using Matlab with the KITTI dataset

I know the folder 'poses.txt' contains the ground truth poses (trajectory) for the first 11 sequences. Each file xx.txt contains an N x 12 table, where N is the number of frames of this sequence. But, what are these 12 parameters? x,y,z, row, pitch, yaw and what?

My second question is if I want to create my own dataset, how can I acquire these poses with IMU?

Solution

  • Each row of the file contains the first 3 rows of a 4x4 homogeneous pose matrix flattened into one line.
    It means that this matrix:

    r11 r12 r13 tx
r21 r22 r23 ty
r31 r32 r33 tz
0   0   0   1

    is represented in the file as a single row:
    r11 r12 r13 tx r21 r22 r23 ty r31 r32 r33 tz
    Search "4x4 homogeneous pose matrix" in Google or read this:
    http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL_COPIES/MARBLE/high/pose/express.htm
    or this
    https://math.stackexchange.com/questions/82602/how-to-find-camera-position-and-rotation-from-a-4x4-matrix