**5. Result Comparison**

The resulting combination (Figure 12) of the point clouds from MMS and CRP surveys was assessed using the cloud-to-cloud (C2C) distance tool. The C2C tool exploits the nearest neighbour algorithm to compute the Euclidean distance between each point of the compared cloud and the nearest point of the reference cloud [32].

**Figure 12.** "Closest points set" tool applied to CRP point cloud, taking the MMS point cloud as a reference.

First, we computed the "closest point set" [34] applied to the CRP point cloud. This tool is useful for understanding which points of the CRP point cloud are closest to each point of the MMS cloud.

Taking the MMS point cloud as a reference, a new CRP point cloud was generated, preserving only the closest points. This new point cloud represents a new base for the C2C distance calculation.

Then, the C2C distance was calculated by taking the resampled CRP point cloud as a reference and setting the point distances in different ranges, from 1.00 to 0.10 m to evaluate different computations in term of RMSE, standard deviation and number of points (Figure 13).

Analyzing Figure 13, it can be noted that setting the distance computation at 1 m between the two point clouds, we have lost almost 8.8 million points with respect to the original one. The former value represents the points that have any correspondence in the CRP point cloud, excluding then the parts of the roads that cannot be seen from the camera due to the occlusions. The threshold of 0.50 m between the two point clouds was computed to understand better the distribution fitting of C2C absolute distance. In the end, we set the C2C distance computation at 0.10 m. The choice of this threshold value is motivated by the fact that long-range MMS and CRP are techniques characterized by sub-decimeter accuracy. The number of points has become low: 17 million points of MMS have a higher correspondence with the CRP point cloud. From the results of the point clouds assessment, it emerges that, at 0.10 m distance, a mean difference of 5.6 cm between laser scanner and photogrammetry exits, with a standard deviation of 2.3 cm (Figure 13).

The reduction of the number of points during the different distance computations can be explained as follows: first, the MMS point cloud is denser than the CRP "closest point set" point cloud, so only a low number of points between the two point clouds has the nearest correspondence; second, the MMS point cloud is noisier. The decrease of the number of points in MMS point clouds during the phases of the C2C evaluation does not imply an information waste, since the remaining 30 million points of MMS point clouds are partially useful to fill the "holes" of the CRP point cloud.

#### *Appl. Sci.* **2020**, *10*, 6831

**Figure 13.** Cloud-to-cloud (C2C) processing between CRP and MMS point clouds at different values of distances: (**a**) 1 m, (**b**) 0.50 m, (**c**) 0.10 m.

A first observation that can be made by seeing the two point clouds, is that the final 3D point cloud (Figure 14) does not present so many holes. The MMS allowed to acquire the elements close to the path followed along the road and to compensate for the gaps that had been created in the CRP point cloud.

This explains that a possible combination of different surveying techniques, their different use and the different data acquired can be complementary and guarantee a complete mapping of the same site of interest. The previous statement is true as far as both the RMSE and the standard deviation are small. Besides the integration between CRP and MMS, we also computed the accuracy evaluation between the aligned MMS point cloud and the photogrammetric control points on natural features. As explained in Section 3.1, the EDM technology system is based on pillars for control and spherical targets for checking the photogrammetric accuracy. Once the quality of the photogrammetric survey has been confirmed, photogrammetry can also be used for checking the accuracy of the MMS based on natural features. Taking the 22 check points of the CRP solution as a reference, we computed the distance on three natural check points (randomly combined) doing a point-pair registration. As shown in Figure 15, the error value is subdecimeter and comparable to the distance evaluation between CRP and MMS point clouds.

**Figure 14.** Integration of filtered point clouds from MMS (Mobile Mapping System) in intensity scale and CRP (Close-Range Photogrammetry) in RGB scale.

**Figure 15.** Point-pair registration between MMS (Mobile Mapping System) point cloud and natural photogrammetric check points.
