Kitti

Understanding the Calibration file

This file basically contains all the transform between between different frames of camera such as velodyne, camera, etc.

Use following links to understand it:

KITTI 3D Object Detection DatasetMedium

Main equation for transformation

y = P_{rect}^i x^{rect}

In the calib.txt file, the $P_0, P_1, P_2, P_3$ matrices are the projection matrix which project the 3D point $x_{rect}$ in rectified camera coordinate frame to image coordinates of camera $i$ . Please note that rectified camera coordinate frame is not related to any camera $i$ . So basically these $P_i$ given in calib.txt are actually $P_{rect}^i$ and takes points in rectifies camera coordinate and not points in $i^{th}$ camera coordinate frames.

y = P_{rect}^i R_{rect}^0 Tr^{velo-to-cam0} x^{velo}

This is one of the main equation given in the readme.md file for transformations of points. Here $x^{velo}$ is 3D point in velodyne coordinate frame. $Tr^{velo-to-cam0}$ transform an point from velodyne to reference camera (cam0) coordinate frame. ie. $x^{cam0} = Tr^{velo-to-cam0}x^{velo}$ . $x^{cam0}$ is in cam0 coordinate frame, cam0 is reference camera in kitti.

x^{rect}= R_{rect}^0x^{cam0}

We need $x$ in rectified frame to be worked with $P_{rect}^i$ . Hence we use $R_{rect}^0$ which transform point from cam0 coordinate frame to rectified coordinate frame.