search

segwayrobotics / segway_DRIVE_benchmark

3 stars 0 forks source link

Some question about configuration files #2

Closed linhuixiao closed 5 years ago

linhuixiao commented 5 years ago

Hello, I test your newly upload sample dataset, and also meet some question.

  1. In https://drive.segwayrobotics.com/#/dataset/index， the Sample Dataset Config file(*.yaml) your provide is show： %YAML:1.0 Author: SegwayRobotics camera_distortion_parameters: w: 8.9130900000000002e-01 camera_projection_parameters: fx: 2.6186799999999999e+02 fy: 2.6019200000000001e+02 cx: 3.2501700000000000e+02 cy: 2.3733500000000001e+02 I just wonder which camera model should be take in the newly upload sample dataset？ PINHOLE or KANNALA_BRANDT ？If using PINHOLE model，the distortion_parameters k1, k2, p1, p2 should be given. If using KANNALA_BRANDT model, the projection_parameters k2~k5,mu,mv,u0,v0 should be given.

  2. In https://drive.segwayrobotics.com/#/overview/sensor_setup， your camera image parameters is： camera 640x480, global shutter, 30fps Camera: 133x100x166.5 FOV @10Hz which image_width and image_height should be taken？which model should be taken? Is fisheye?

qinyusen commented 5 years ago

Hi,

For the 1st question, I will give a short explain, from very beginning, since maybe you mix up the camera models.

There are 2 definitions, projection models and distortion models.

For us we are using pinhole as projection model, and FOV as distortion modle. Some projection model:

  1. pinhole camera model (intrinsics vector: [fu fv pu pv])
  2. omnidirectional camera model (intrinsics vector: [xi fu fv pu pv])
  3. double sphere camera model (intrinsics vector: [xi alpha fu fv pu pv])
  4. extended unified camera model (intrinsics vector: [alpha beta fu fv pu pv])

Normally we could find these distortion model:

  1. radial-tangential (distortion_coeffs: [k1 k2 r1 r2]) (This maybe what you are talking about PINHOLE modle)
  2. equidistant (distortion_coeffs: [k1 k2 k3 k4]) (This is what you are talking about "KANNALA_BRANDT model")
  3. FOV (distortion_coeffs: [w]) (This is what we are using here.)

For the FOV model, you can refer on this paper: https://hal.inria.fr/inria-00267247/file/distcalib.pdf

And you can find some similar information on : https://github.com/ethz-asl/kalibr/wiki/supported-models

For the 2nd question:

"camera 640x480, global shutter, 30fps" this means the resolution is 640x480 ,which is the image width and height, and running under 30fps

Camera: 133x100x166.5 FOV @10hz FOV is the field of vision, 133x100x166.5 means the horizontal FOV is 133 degree, vertical FOV is 100 degree, and diagonal FOV is 166.5 degree.

Thanks.

qinyusen commented 5 years ago

You can try with this modified project : https://github.com/pangfumin/VINS-Mono

check the readme 3.14

qinyusen commented 5 years ago

@linhuixiao Do you have any further question?