关于KITTI标定文件参数格式及里程计标定文件的技术问询

Hey folks, let's dive into understanding the KITTI calibration file format and parse the snippet you shared. I'll break this down clearly so you can make sense of every part.

KITTI Calibration File Format Basics #

First, let's cover the core rules of how KITTI structures its calibration data:

• Each calibration parameter starts with a label like P0: or Tr:, followed by space-separated floating-point values (in scientific notation here). These values represent matrix elements in row-major order.
• Key parameters you'll typically see include:
  • P0/P1/P2/P3: 3x4 projection matrices for the four KITTI cameras. These map 3D points in camera coordinates to 2D image pixels.
  • R0_rect: 3x3 rotation matrix used to rectify the camera images (corrects for lens distortion and camera pose misalignment).
  • Tr: 4x4 transformation matrix that converts points from LiDAR coordinate system to the reference camera's coordinate system.

Parsing Your Provided Calibration Snippet #

Let's unpack the P0 and P1 data you shared:

P0 Projection Matrix #

The P0 line translates to a 3x4 matrix (row-major order) as follows:

718.856    0        607.1928   0
0          718.856  185.2157   0
0          0        1          0

Here's what each component means:

• Focal length: fx = fy = 718.856 (symmetric, typical for pinhole cameras)
• Principal point: cx = 607.1928, cy = 185.2157 (the center of projection in image pixels)
• The last column is all zeros, which tells us this is the reference camera (no translational offset relative to the calibration origin).

P1 Projection Matrix #

The partial P1 line you provided gives us the first two full rows and part of the third. Arranged as a matrix (so far):

718.856    0        607.1928   -386.1448
0          718.856  185.2157   0
... (remaining elements omitted)

Key observations here:

• Same focal length and principal point as P0, which makes sense since KITTI uses identical stereo cameras.
• The fourth element in the first row is -386.1448 — this is the horizontal translational offset between the left (P0) and right (P1) cameras, a critical value for stereo depth estimation.

内容的提问来源于stack exchange，提问作者Finley