Issue of scale in the final fused point cloud


I have placed a ZED2i -4mm cam on a vehicle that had a speed of about 30 km/h. The final point cloud is apparently smaller by about 4%. It means the distances are about 96% of the ground truth values. I tried to calibrate the cam using your software based on several medium to x-large screens (i.e., in an auditorium) but the produced config file did not solve the issue, although it slightly alleviated the issue. My questions:

1- After calibration the baseline is some strange value completely different from 120 mm, for example 17 mm. I replaced it with 120 in the config file. Is that correct to do so?

2- How accurately the left and right cameras are synchronized to take the images at the same time? In the case of movement with the speed of 30 km/h, a delay of for example 1 millisecond can explain the scale problem because it changes the length of baseline. I wonder what trigger mechanism is used to ensure a very accurate synchronization between the two image captures from the left and right cameras. If you have any estimate of the possible delay, I would be grateful if you could provide it for me.


Hi @mostafa.hoseini

This is not expected. We are checking why the calibration tool does not correctly retrieve the DPI information from the display settings.
Setting the value manually cannot generate good results, I recommend using the OpenCV calibration instead, while waiting for the fix:

We understand the significance of synchronizing CMOS sensors for a stereo camera. All our devices are designed to trigger the CMOS sensors using the same electronic signal, ensuring perfect synchronization with a precision of microseconds.
Furthermore, when using GMSL2 devices, the Stereolabs GMSL2 capture cards use the same hardware triggering signal for all the sensors of all the cameras connected to them, this guarantees perfect hardware synchronization of all the sensors, for example for all the 4 ZED X stereo cameras connected to a ZED Link Quad.