It is not a good idea to make a group-representative full-cortex map of human data in volumetric format, volume registration doesn't deal with the magnitude of folding variability involved, see our paper:
We think cortical tractography that uses surfaces for detecting cortical hits can generally be better than using a voxel mask, due to the stairstep nature of binary voxel masks. However, if the tool you want to use doesn't support surfaces, what you can do is map a cortical parcellation into each individual's volume space separately, using that individual's own surfaces. See wb_command -cifti-separate, and -label-to-volume-mapping with the -ribbon-constrained option.
Tim