A Digital Terrain Model describes the reconstructed 3D terrain in a suitable geographic coordinate system. Unlike the Surface Model (see DSM), it should contain only points that belong to the terrain layer. This means that it’s implicitly or explicitly performing a crude segmentation of the scene between terrain vs non-terrain points.
In the case of Pix4Dmapper, the algorithm uses the classes terrain (ground), vegetation, building and unknown to locally adapt the DTM extraction process.
Looking at the area, you presented in the link, that medium height grass should be assigned to the vegetation class. Afterwards, the software will use the mild smoothing over that class and get to the terrain.
As the algorithm to differentiate the classes uses geometry and colour characteristic of the objects, you would need to ensure that your point cloud is properly classified (as the area you are planning to reconstruct is very uniform).
How to generate the point cloud classification
How to edit the point cloud in the rayCloud