Wierd results on head-motion estimation in fMRIprep

Dear all,

I processed fMRI data using fMRIprep version 1.5.7 and completed the preprocess without any error.
But I found motion parameters of some participants strange.
The below are the pictures of the three parameters (pitch, roll, yaw, respectively).

fmriprep_pitch fmriprep_roll fmriprep_yaw

The value of roll and yaw is zero for the first about 200 volumes.
I also estimated the parameters using SPM and FSL (MCFLIRT).

spm_pitch spm_roll spm_yaw

fsl_pitch fsl_roll fsl_yaw

From just visual inspection, I guess only fMRIPrep outputs are strange.
I also tried the latest version (20.2.0), but the result was the same.
Can anyone think of anything that might have caused it?