Thanks for the great work.
I am trying to calibrate my multi-camera fisheye(190deg) setup for an autonomous vehicle.
When I try multi-cam approach, I failed due to lack of co-obervations. Since the system is intended for 360deg surround view, there is little overlap between cameras.
Later, I tried to calibrate each camera within an individual rosbag. So, I record a rosbag for each camera. By this approach, with pinhole-equidistant model, I have the re-projection errors around 0.15-0.25 (average). Do you think the values are ok? (I consider to use ros-validator to get a better feeling.)
One final question regarding the how image size affects the calibration results. My original image res. is 1920/1280, I have to downsample images to the 640x480 for faster computation, so I do all my calibration efforts in 640x480. Would it be affecting calibration performance too?
Hello everybody,
Thanks for the great work. I am trying to calibrate my multi-camera fisheye(190deg) setup for an autonomous vehicle. When I try multi-cam approach, I failed due to lack of co-obervations. Since the system is intended for 360deg surround view, there is little overlap between cameras.
Later, I tried to calibrate each camera within an individual rosbag. So, I record a rosbag for each camera. By this approach, with pinhole-equidistant model, I have the re-projection errors around 0.15-0.25 (average). Do you think the values are ok? (I consider to use ros-validator to get a better feeling.)
One final question regarding the how image size affects the calibration results. My original image res. is 1920/1280, I have to downsample images to the 640x480 for faster computation, so I do all my calibration efforts in 640x480. Would it be affecting calibration performance too?
thanks in advance!