search

MARSLab-UMN / MARS-VINS

MARS-VINS binary release repository
http://mars.cs.umn.edu/
Other
116 stars 41 forks source link

camera calibration (R,t) #10

Open protobits opened 6 years ago

protobits commented 6 years ago

Hello, I'm trying to convert the dataset to rosbag format and I have come upon a problem with the extrinsic camera calibration. It seems that the relative transform between left and right camera results in the right camera a bit behind the left camera. Is that correct?

image

I'm using the parameters from the example config file where this transform is already computed, but I think it is the same as the one expressed in the dataset config, only w.r.t. the imu.

I'm asking since when looking at the pictures of the camera rig in the paper it seems difficult for one camera to be behind the other. Moreover, when performing stereo-rectification I'm not getting straight epipolar lines.

goldbattle commented 6 years ago

Normally the z-axis would be the front of the camera. To me, this shows that the left camera is to the left of the right, and a bit more forward.

protobits commented 6 years ago

Actually, this picture was made using the left-to-right camera transform found on the example config. If I plot the same thing using the transforms w.r.t. to the camera I get the left one at the right. Anyway, in both cases I get the same 2cm difference in depth of the cameras, which seems wrong looking at the pictures of the camera rig in the paper. Moreover, rectification with these parameters is wrong.